docs(res-to-affine): corpus run + regex precision fixes (Refs #57)

[hyperpolymath]

Summary

First end-to-end run of the Phase-1 scanner (#314) against the estate's
real .res surface — 491 deduplicated files across idaptik (475) and
gitbot-fleet (16). The run surfaced two high-impact false-positive
sources, which this PR fixes alongside the corpus report.

• side-effect-import was flagging in-function let _ = X.f(...)
  — ReScript's normal "discard chained-call return value" idiom — not
  just the LESSONS.md "module-load side effect" anti-pattern. 1,181
  hits in the corpus, vast majority indented.
• mutable-global was flagging any :=, including
  intra-function local-ref mutation. 653 hits in the corpus, none at
  column 0.

Both regexes now anchor at column 0. re_untyped_exn also picks up
column-0 raise / try via a (^|...) alternation.

Impact

	Before	After	Δ
Total markers	2,146	348	−84%
Files with markers	216	94	−122
side-effect-import	1,181	36	−1,145
mutable-global	653	0	−653
raw-js	198	198	unchanged
untyped-exception	114	114	unchanged

Spot-checked the new top hotspots — every remaining side-effect-import
hit is genuinely column-0 (e.g. idaptik/src/Main.res:5:let _ = PixiSound.sound);
every raw-js and untyped-exception hit corresponds to a real %raw(…)
block or try/Js.Exn/Promise.catch occurrence.

What's added

• tools/res-to-affine/CORPUS-RUN.md — human report (corpus stats,
  regex change-log, top hot-spots, Phase-2 follow-ups, reproducer).
• tools/res-to-affine/CORPUS-RUN.json — machine-readable sidecar.
• tools/res-to-affine/README.md — link to both.
• tools/res-to-affine/scanner.ml — three regex tightenings with
  an inline comment naming the column-0 vs in-function trade-off.

Trade-off

We no longer flag:

• module-load side effects nested inside module X = { ... } blocks
• top-level let x = ref(...) declarations (we still flag the :=
  that follows, if column-0)

Both are scoped to Phase 2's AST walker and listed in
CORPUS-RUN.md as follow-ups (item 2 + item 3).

Refs #57 (the issue is multi-phase, this is a precision pass on
Phase 1, not a phase close).

Test plan

• dune build tools/res-to-affine — clean.
• dune test tools/res-to-affine/ — 3/3 OK (snapshot byte-identical
  because the synthetic fixture covers column-0 cases only).
• Re-ran scanner over the 491-file corpus, verified the per-kind
  tally matches the table above and spot-checked remaining hits.
• CI on this PR.

🤖 Generated with Claude Code