JS WAM: distinct string term tag

[s243a]

G-W3 residual: the JS WAM now has a distinct string tag (V.String) instead of interning SWI strings as atoms.

Branched from grok/wamjs-op3 and extended it (net additions). No INTEGRATION_PATCH.md.

What landed

• Term model: V.String(s) alongside atom/int/float/var/struct. deref, unify (string unifies only with an equal string, never an atom of the same text), copy_term, ==/\==, and compare/3 / sort handle the tag.
• Standard order matches SWI 9.0.4 (the oracle on this tree): Var < Number < String < Atom < Compound. "foo" @< foo / compare(foo, "foo") is >. The older “Atom < String” manual line is not this SWI version.
• atom/1 is false for strings; string/1 is true only for the tag.
• String-producing builtins now box V.String:
  • atom_string/2
  • string_concat/3 (atom_concat still produces atoms)
  • string_chars/2 construct (decompose is still a list of one-char atoms)
  • string_to_atom/2
  • number_string/2
  • split_string/4 (list of strings)
• write/1 prints the text without quotes (SWI write). format ~q quotes with ".
• Pratt reader tokenizes "..." as the string tag.

Compiled constants stay atoms. The shared WAM tokeniser stores every constant as a Prolog string token, so wrapping put_constant in V.String would rewrite atoms/ints and break the 48-query suite. Construct strings via builtins or the Pratt "..." reader.

Tests

probe_string_tag/0 in tests/test_wam_javascript_builtins.pl:

• atom_string(a, S), string(S), \+ atom(S)
• split_string('a,b,c', ',', '', Parts) is a list of strings
• string_concat(x, y, Z) yields a string
• sort([foo, SFoo, 1, bar], L) → [1, SFoo, bar, foo]

Local acceptance:

1. tests/test_wam_javascript_builtins.pl green (incl. new probe)
2. tests/test_wam_javascript_lowered.pl and tests/test_wam_javascript_fact_sources.pl green
3. Official CONFORMANCE_TARGETS=javascript harness exits 0 (adapter still unpatched; local 48-query stand-in in the builtins file stays 48/48)
4. SWI vs Node dumps match for string/1, atom_string, split_string, string_concat, and standard-order sort

Residuals

• writeq/1 is not registered as a standalone builtin (quoted rendering is via format ~q)
• Compiled "foo" literals collapse to atom foo (shared WAM constant tokens; cannot fix without editing wam_target.pl)
• Fact-source JSON/TSV cells still intern as atoms
• string_length/2 not added
• format('~q', [List]) does not quote nested strings (list walk uses write, not writeq)
• Cyclic terms are N/A for the string tag itself