the measurement records what the declaration did, and a refusal is not a failure

The pinned graph now declares [c-abi], and the Windows leg went from
15 runs / 15 fails to 23 runs / 7 fails. Linux is unchanged at 27 / 3.

Eight members that had failed on a missing Windows header run unchanged:
asio (through cmp-module), catch2, cli11, eigen, fmtlib.fmt, libpng, re2 and
lua (through capi-lua). Each selected its Windows branch on `_WIN32` or
`__MINGW32__`; with neither defined it takes the POSIX branch it takes on
Linux, and needs nothing in its descriptor. This is the first measurement in
which the declaration was in effect at all.

The seven that remain are three groups, and only the first is about the C
environment. mimalloc and sqlite3 stop on `windows.h` because `__CYGWIN__` is
defined and each reads that name as "the Win32 API is available" --- mimalloc's
own comment says "we use windows locks on cygwin". That settles half of a
trade-off mcpp's documentation had left explicitly open to this measurement: a
name borrowed from another environment carries the meaning its lender gave it.
archive, c-ares and curl want a generated configuration, as they do on every
system. doctest and spdlog are new and are progress: they used to stop at a
missing header and now reach the link, naming `__cxa_thread_atexit` --- a gap
in the C++ runtime above openkal, not reachable until the environment was
right.

`refused` is added as a status. A member told before compilation that this
graph does not supply a capability it asked for has been answered correctly,
and counting that as a failure turns the compatibility figure into a score the
build tool is judged by --- which is how a build tool ends up simulating an
operating system it is not running on. The judge is mcpp's own refusal code
and not a string in the diagnostic: the member declared the requirement and
the graph answered. No member carries it today, and the summary counts it
apart from `fails`.

Co-authored-by: Claude Code <noreply@anthropic.com>

Author: Sunrisepeak

.github/workflows/openkal-compat.yml

@@ -131,6 +131,23 @@ jobs:
           python3 - <<'PY' >> "$GITHUB_STEP_SUMMARY"
           import json
           r = json.load(open("openkal-compat.json"))
+          # Counted per target, and `refused` counted apart from `fails`: a
+          # member told before compilation that this graph does not supply a
+          # capability it asked for has been answered correctly, and adding it
+          # to a failure figure turns that figure into a score the build tool
+          # is judged by.
+          tally = {}
+          for e in r["members"].values():
+              for t, rec in e["targets"].items():
+                  tally.setdefault(t, {}).setdefault(rec["status"], 0)
+                  tally[t][rec["status"]] += 1
+          print("| target | runs | builds | fails | refused |")
+          print("| --- | --- | --- | --- | --- |")
+          for t in sorted(tally):
+              c = tally[t]
+              print(f"| {t} | {c.get('runs',0)} | {c.get('builds',0)} "
+                    f"| {c.get('fails',0)} | {c.get('refused',0)} |")
+          print()
           print("| member | target | result | first diagnostic |")
           print("| --- | --- | --- | --- |")
           for m, e in sorted(r["members"].items()):