docs: transcribe #943's CHANGELOG entry and add its v3.0.0 SemVer break

[gsdali]

What & why

PR #944 (#943) merged without its CHANGELOG entry and without its SemVer break row. This adds both.

• The CHANGELOG entry existed in fix(#943): make Shape.bounds Optional for v3.0.0 breaking change #944's PR body and was never transcribed. check-changelog-transcription.py --verify-transcribed on main reports it: 882a2ddb #944 under "entry in the PR body, NOT in the CHANGELOG".
• The SemVer row could not have been in fix(#943): make Shape.bounds Optional for v3.0.0 breaking change #944, because semver-at-release forbids a PR touching docs/SEMVER.md. Writing it is the merging step, and it was missed.

What would have shipped

docs/SEMVER.md's v3.0.0 break table listed only #844 — an enum rename and a RawValue change. Meanwhile six accessors had become Optional on main:

Shape.bounds, Shape.size, Shape.center, Wire.bounds, Edge.bounds, Face.bounds

That is a compile error for every caller, and it is the break a consumer is most likely to hit, because bounds, size and center get read casually.

This is #829 repeating. v2.0.0 shipped with #595 missing from its break table; a real downstream consumer (swiftGCS) hit a compile error that reading the table could not have predicted; we corrected it four days ago in PR #939. Same gap, one release later, same cause: a break whose row nobody wrote at merge time.

Changes

• docs/CHANGELOG.md: Breaking: Shape.bounds becomes Optional, so a void shape stops fabricating (0,0,0)-(0,0,0) (follow-up to #834) #943's entry, transcribed from fix(#943): make Shape.bounds Optional for v3.0.0 breaking change #944's body and expanded slightly with the root cause the PR established — occtComputeBoundingBox being file-static in OCCTBridge_Topology.mm is why OCCTShapeGetBounds, over in OCCTBridge_Properties.mm, was the one bounds entry point with no guard.
• docs/SEMVER.md: a third row in the v3.0.0 table, and a ##### v3.0.0: bounds, size and center become Optional (#943) subsection with the full before/after table and a migration snippet.
• The v3.0.0 intro said the major was "carried by a single merged change". It now names both.

The subsection records the measured reason a value-based test was rejected: a vertex at the world origin measures exactly (0,0,0)-(0,0,0) through BRepBndLib::AddOptimal, so inferring "void" from the coordinates is the same fabrication relocated. Through ordinary BRepBndLib::Add the same vertex measures ±1e-7, because BRep_Tool::Tolerance floors at Precision::Confusion() — so a sentinel would have misfired on one path and not the other. Both from Scripts/repro/943-bounds-void-vs-zero/.

Verification

All seven gate scripts run clean, including check-changelog-transcription.py, which no longer reports #944.

CHANGELOG entry

None. This PR is the transcription; per changelog-on-merge a PR whose only purpose is to write an entry from another PR's body is the merger doing their job, and it is the only open PR touching either file.

SemVer impact

NONE in itself. Documentation only. It documents a break that already landed in #944; it does not introduce one.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (2 files)

• docs/CHANGELOG.md - Added missing CHANGELOG entry for Breaking: Shape.bounds becomes Optional, so a void shape stops fabricating (0,0,0)-(0,0,0) (follow-up to #834) #943 (bounds/size/center becoming Optional)
• docs/SEMVER.md - Added v3.0.0 break table row and detailed subsection for the same change

---

_{Reviewed by nemotron-3-ultra-550b-a55b:free · Input: 218K · Output: 1.7K · Cached: 336.6K}