Avoid repeated landmark photo preparation

[timohueser]

The landmark capture repeated the full offline content compile after every photo fallback. A country capture therefore resized and wrote every settled photo again when only a small set of QIDs could change.

This adds a request-only compiler result. The first round evaluates all QIDs. Each later round evaluates only the QIDs requested in the preceding round. The compiler still reads and validates the candidate bytes before it chooses a fallback. It emits only photo-requests.json; ordinary content compilation keeps its existing output.

The frozen Q176775 fallback case produced the same full and filtered request document. A release build from this branch and a release build from origin/develop produced identical records, counts, input digest, category policy digest, and RGB222 bytes. Only policy_sha256 differs, as expected because the compiler source changed.

Checks:

• python3 -m unittest tools.tests.test_landmark_capture (29 passed)
• obc test -p obc-pack (463 passed)
• obc test -p obc-bake (96 passed)
• cargo clippy -p obc-pack -p obc-bake --all-targets -- -D warnings
• obc suites check
• cargo build --release --locked -p obc-bake
• Frozen release comparison against origin/develop

Requirements: none

Public docs: no

Closes #2162
Relates #2095
Relates #1978

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 5d22aadf-1fe0-4684-8262-56dc919c7965

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[timohueser]

Independent GPT-5.6 Sol review accepts bca69f52f8415a23e392019acc4fac870a6d110e with no remaining findings.

The first review required a real production-path equivalence test. The test-only delta now covers a settled usable photo, a rejected original and missing fallback, and an unrelated QID. It checks full request order, exact filtered/full request equality, fallback identity, and request-only output isolation. The existing Python test covers the acquisition loop and independent per-QID path updates for a shared source.

The implementation preserves source validation, fallback decisions, input immutability and normal content behavior. The compiler policy fingerprint changes as expected from the source change. Selected CI remains the merge gate. Adoption by the retained Swiss capture needs a separate full-corpus compatibility and provenance check.