various fixes

[dcolazin]

In this PR I collected several fixes required for my usage on Linux.

Bug 1 — No bestmove output when there is no mate solution

Problem: The UCI protocol requires that every go command be answered with exactly one bestmove line. When chest reported No Solution, extract_bestmove() printed nothing, hanging the GUI.

Root cause: The bestmove buffer stayed empty because the move-parsing loop was guarded by if (mate > 0) and the final print was conditional on bestmove[0] being non-null.

Resolution:

• Added a parse_refu_move() helper that extracts the first legal move from chest's refu (refutation) lines — the moves that refute the mate attempt are valid legal moves.
• When check_no_solution() triggers, the code now scans all lines for a refu entry and uses the first refutation move as the bestmove.
• The final output is now unconditional: if a bestmove was found (either from the solution or from a refutation) it is printed; otherwise the code emits bestmove 0000 (the UCI null-move), guaranteeing a response in all cases.
• If the output file cannot be opened at all, bestmove 0000 is emitted immediately.

---

Bug 2 — backend setoption truncates the last character of the path

Problem: Setting the backend path via setoption name backend value ./dchest corrupted the path. Two separate defects combined to eat characters: v += 7 skipped one character too many (should have been v += 6), and an unconditional dchest[strlen(dchest) - 1] = '\0' stripped a valid trailing character under the incorrect assumption that a " quote was present.

Resolution:

• The entire setoption handling block was rewritten into a unified parser that accepts both the full UCI form (setoption name <key> value <val>) and a short form (setoption <key> <val>).
• The value keyword offset is now correct (exactly 6 characters consumed after detection).
• Quote stripping is now conditional: a trailing " is removed only if one is actually present, and a leading " is removed only if one is present. Unquoted values (the normal UCI case) pass through untouched.

---

Bug 3 — position startpos is not handled

Problem: The adapter only handled position fen .... When a GUI sent position startpos or position startpos moves ..., the internal FEN buffer was left empty, producing garbage output from chest.

Resolution:

• Added a START_FEN constant (rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1).
• A new handler (!strncmp(line, "position startpos", 17)) sets current_fen to START_FEN before the existing position fen handler, so the standard starting position is used.

---

Bug 4 — parse_chest_move produces malformed UCI moves

Problem: The original parser naively stripped the first uppercase letter and kept every remaining alphanumeric character. This mangled captures (Qxg6 → xg6), promotions (e8=Q → e8Q), and castling (O-O → O). The only formats that happened to work were the spaced/colon-separated notations (Ra1 - a8, Rf6 : f7).

Resolution:
The function was completely rewritten with a multi-step pipeline:

1. Normalization — The entire line is copied into a buffer with all whitespace, separators (-, :), and check/mate suffixes (+, #) removed.
2. Castling — If the token starts with O or 0, the O-count determines kingside (OO → e1g1) or queenside (OOO → e1c1).
3. Piece-letter stripping — A single leading uppercase piece letter (non-O) is skipped.
4. Promotion — If = is found, the following piece letter is captured and lowercased as a UCI promotion suffix.
5. Capture cleanup — Any x or X characters are removed.
6. Assembly — The remaining clean token is assembled into UCI coordinate notation, appending the promotion suffix when applicable.

---

Bug 5 — go without depth is not handled

Problem: Only go depth N was recognised. Any other go variant (go movetime 1000, go wtime 5000 btime 5000, bare go) was silently ignored, hanging the GUI.

Resolution:

• The handler now matches any line starting with go.
• It searches for a depth substring; if found, the depth value is extracted as before.
• If no depth is specified, a default depth of 3 is used.
• In both cases write_chest_input() and run_chest() are called, so the engine always responds.

---

Bug 6 — Dead variable bestmove in main()

Problem: A char bestmove[32] local in main() was never used (the real best-move logic lives in extract_bestmove()). It generated a compiler warning under -Wall -Wextra.

Resolution: The declaration was removed.

---

Bug 7 — Fragile fpos_t file positioning

Problem: The original two-pass approach used fpos_t / fgetpos / fsetpos to save and restore the file stream position. This was fragile: fpos_t is an opaque type, the initial = 0 initializer caused a compilation error on glibc (see Bug 8 below), and fsetpos was called unconditionally — potentially on an uninitialized setPoint when no mate solution was found.

Resolution:

• The entire output file is now read into an in-memory array of line strings (lines_buf) using strdup().
• Both passes iterate over the in-memory array, eliminating all fpos_t / fgetpos / fsetpos usage.
• The file is closed immediately after reading.
• All allocated line buffers are freed at the end of extract_bestmove().

---

Bug 8 — Compilation error: invalid fpos_t initializer

Problem: fpos_t setPoint = 0; failed to compile with gcc -O2 because fpos_t is an opaque type (a struct under glibc with large-file support) and cannot be initialized with an integer literal.

Resolution: As part of the Bug 7 refactor, the fpos_t variable and all associated calls were removed entirely, resolving the compilation error at its source.

---

Bug 9 — Unchecked system() return value

Problem: system(cmd) discarded its return value, triggering a warn_unused_result warning on glibc and making it impossible to detect a failed backend launch.

Resolution: The return value is now checked; on failure a Warning: failed to run chest command message is printed to stderr.

---

Additional Improvement — stdin redirection for dchest

Change: The run_chest() command was changed from "%s %s > %s" to "%s %s < /dev/null > %s".

Rationale: Redirecting stdin from /dev/null prevents dchest from accidentally reading from the UCI adapter's input stream, which could consume UCI protocol commands intended for chest_uci.

---

Additional Improvement — Selfmate / helpmate score negation

Change: When job_type is s (selfmate) or h (helpmate), the score mate value reported in the info line is negated.

Rationale: In these problem types the side to move is the one being mated, not delivering mate. Without negation, GUIs like ChessX would display a selfmate solution as if it were a direct mate. A negative mate score correctly communicates "the side to move will be mated in N."

---

Additional Improvement — Default job_type

Change: job_type was changed from an empty string ("") to "o" (orthodox).

Rationale: Ensures a sane default when no setoption name job command has been received, so dchest operates in standard direct-mate mode out of the box.

---