extradoc: keep exact-text pre-pins immovable in table-flank pinning

[cassus]

Closes #75.

Context & disclosure. I hit this bug personally while using reconcile_v3
in SUGGEST mode, and I've personally verified that this fix resolves it on a
real document. That said, the investigation, the fix, and most of this write-up
are largely AI-generated (Claude Opus) — I found the bug and confirmed the fix,
but the prose and code are largely AI-authored, so please weight the detail
accordingly.

This surfaced specifically through SUGGEST mode, which isn't generally
available yet and whose main PR hasn't landed — so it's completely fine if you'd
rather not merge this now. I'm opening it mainly to share the finding upstream
so anyone who hits the same thing can benefit, merged or not. I also haven't
tested whether the same bug (or fix) manifests in direct edits without
SUGGEST mode.

This project has been really useful to me and I'd like to give back: if you want
more human attention, deeper testing, or changes to make this mergeable, I'm
happy to engage personally and put in the work — just say the word.

What

_pin_table_flanks (diffmerge/content_align.py) rebuilds its anchor set from
each matched table's ±1 flanking positions with no awareness of the exact-text
pre-pins already established by _pre_pin_stable_anchors. A pre-pin is a
proven unique correspondence; a flank pin is only a heuristic. When a table's
neighbour shifts (e.g. a new heading inserted adjacent to the table), a flank
pin can land on a base index already held by a correct exact-text pre-pin and
silently overwrite it — so a low-confidence heuristic overrides a
high-confidence anchor, and the displaced paragraph gets duplicated/garbled.

See #75 for the full root-cause and the align_content() pipeline ordering.

Changes

• diffmerge/content_align.py — thread pre_pins into _pin_table_flanks and
  seed them into the anchor set at top priority; the existing conflict-
  resolution loop now only ever drops a flank pin, never a pre-pin.
• tests/diffmerge/test_table_flank_pinning.py — two prior tests that asserted
  the old flank-override behaviour updated to the correct exact-text match;
  adds test_flank_pin_never_overrides_exact_text_pre_pin (fails before the
  fix, passes after).

Verification

• Full unit suite green.
• A/B on a real document overlay: fix reverted → heading re-inserted as a
  duplicate, mangled into the new heading's text, with spurious deletes; fix
  restored → heading untouched, no spurious deletes.
• Real-API SUGGEST push + verify_accepted.py: 227/227 paragraphs clean on two
  independent copies.