Make the fetchall guard line-number-independent (stop the recurring baseline drift)

[Scottcjn]

Problem

scripts/check_fetchall.sh / tests/test_fetchall_guard.py record each legacy raw .fetchall() site in scripts/baselines/fetchall_existing.txt keyed by file:line:content. The line number makes the baseline fragile: any PR that adds/removes lines in node/rustchain_v2_integrated_v2.2.1_rip200.py above a baselined call shifts that call's line number, so it no longer matches its baseline entry and the guard reports it as a 'new unannotated' site — turning the blocking test check RED on main and on every open PR.

This has now required manual resyncs twice in one day (#6870, then #6871 after #6866 shifted lines again). It's whack-a-mole.

Proposed fix

Key the baseline by content, not file:line:

• Store file:normalized_content (drop the line number) in the baseline.
• In the guard, transform both the scan and the baseline the same way before comparing.
• To preserve detection of a newly-added call whose content duplicates an existing one, compare as a multiset (e.g. sort | uniq -c counts per file:content) rather than sort -u, so adding an Nth identical-content call still trips the guard.

Acceptance

• Adding/removing unrelated lines in a node file does NOT change the guard result (no false 'new' from line shifts).
• A genuinely-new .fetchall() (new content, or an additional instance of existing content) IS still flagged.
• tests/test_fetchall_guard.py updated to cover: (a) a line-shift produces no error, (b) a new distinct call is caught, (c) a duplicate-content new call is caught.
• Migrate scripts/baselines/fetchall_existing.txt to the new format.

Why it matters

This is the same class of recurring CI tax as the miner-checksum manifest (fixed structurally by the auto-regen hook #6857) and the clawrtc flaky test (#6863). A line-independent guard ends the resync treadmill for good.

Filed after resyncing the baseline twice (#6870, #6871) in one session.

[sayuru-akash]

Submitted a focused fix in PR #6880: #6880

Scope:

• baseline entries now compare as file:normalized_content instead of file:line:content
• duplicate normalized entries are kept as a sorted multiset, so an added identical .fetchall() call still trips the guard
• migrated scripts/baselines/fetchall_existing.txt to the line-independent format
• added tests for line-shift tolerance and duplicate-content detection

Validation available from this environment: bash -n for the updated guard and py_compile for the updated test file. Full repo checkout stalled in index-pack, so I could not run the full pytest locally.

[rebel117]

I'd like to take a stab at this. The line-number fragility is a real pain point — I'll rework the baseline to key on file:content instead of file:line:content and update both the check script and the test.