Let PR validation grade a failing file instead of dying on it

[andrewboudreau]

The gate dies exactly when it has something to say

Since #1367 the PR validator has been unable to report on any PR containing a
non-reproducing file. The draft-banner rule reads the module under a name the import
does not bind:

import asm_policy as AP          # line 49
...
if w == "NO-REPRO" and asm_policy.has_draft_banner(text):   # line 342

and short-circuits, so the line is only evaluated once a file grades NO-REPRO —
and then it raises NameError in the middle of the classification loop. The worker
reports

worker error: pr_linkcheck exited 1 without a report

with no verdict for any file, including the ones it had already graded. #1390 is
the first PR to hit it.

What it was hiding

Reproduced locally against #1390's head with the fix applied — 310 files:

verdict	count
VERIFIED	305
BLIND	3
DRAFT	1
NO-SYM	1

The crash file is src/_ZN3MrI13InitResourcesEv.cpp, which sorts immediately after
src/_ZN3Amp8BehaviorEv.cpp — the last line the validator printed before dying. It
carries a // NONMATCHING banner, so the correct verdict is DRAFT, a pass. The
rule was written for exactly this file and had never once run.

The NO-SYM is real and pre-existing, not from #1390:
src/_ZN6Player16CleanupResourcesEv.cpp does not compile on clean main either
(identical error, and the PR does not touch it). That is #1367's NO-SYM category doing
its job on the first PR whose fan-out reached the file. Fixed separately in
fix/player-cleanupresources-compiles.

The fix

One word — plus the reason it survived. The rule is only reachable when a file
fails, which is the one path a green CI run never exercises, so sitting inline in
main() it was untestable and untested. The two source-text overrides move into
source_policy(worst, text) and tools/test_pr_linkcheck_verdict.py drives them
directly, including the collision the old inline comment asserted could not happen (a
bannered transcription grades DRAFT, not RAW-ASM).

Behaviour is otherwise unchanged: the extracted function preserves the original order,
and a draft banner still does not excuse WRONG.

Verified

• tools/test_pr_linkcheck_verdict.py — 7 pass; test_pr_linkcheck_renames.py and
  test_asm_policy.py still pass (24).
• pyflakes tools/*.py in both directions: one undefined name on main
  (pr_linkcheck.py:342), zero across all of tools/ with this applied.
• pr_linkcheck.py re-run end-to-end on TextureSequence::Prepare and MaterialChanger::Prepare are static #1390's head: completes, 310 files graded, no
  crash.
• No src/ or config/ change, so the ROM build is untouched.

There is no Python lint step in CI today. Adding one would have caught this on the
commit that introduced it, and the baseline is already clean — but that is a separate
change, not folded in here.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XxDmkQ47fWa3GB6mj8xEQe

[tangos-validator]

✅ PR validation — Passed

noverify: no source/build-data changes in this PR

_{Each changed src/*.c|*.cpp is compiled and its relocated bytes compared to the binary data on a private build box. Passing requires every changed file to reproduce the ROM byte-for-byte with correct relocation targets — this catches WRONG-DEST relocations and non-reproducing near-misses that ledger-scoped linkcheck skips.}