ps-plate-04-refine: attitude re-fit, RA/Dec/Roll, FOV/distortion refinement, residuals, and solution outputs

[bryantharpe]

Implements the refinement stage of the lost-in-space plate solver.

• Re-fits attitude over all matched pairs with the SVD Wahba solver.
• Extracts RA/Dec/Roll from the refined rotation matrix.
• Refines FOV and optionally radial distortion from matched pairs.
• Computes residual statistics (RMSE/P90E/MAXE) in arcseconds.
• Assembles a fully populated Solution including camera, matches, and status.
• Adds VerificationOutcome to carry matched pairs from verification to refinement.
• Wires refine_solution into the solve loop on first accepted candidate.

All existing tests pass.

[bryantharpe]

review/judge — APPROVE on 0a48c92d0aac

VERDICT: APPROVE
BLOCKING: NONE
NITS: crates/plate-solver/src/refine.rs:22 — parameter named _candidate is actually used (vec![*_candidate]); drop the underscore prefix to avoid implying it is unused.
NITS: crates/plate-solver/src/verify.rs:52 — doc comment says "Returns None if the candidate is rejected before the projection/match stage" but None is also returned when the binomial test rejects after matching; consider rewording to "Returns None if the candidate is rejected at any stage."
NITS: crates/plate-solver/src/refine.rs:82 — refine_fov(...).0 discards a second return value without comment; a named binding or brief comment would clarify what is being dropped.

---

Independent U3 review by the judge role (glm-5.2, Zhipu — different lineage
than every author seat). Verdict is on blocking findings, not a score. To
supersede with human sign-off (U4): judge-pr.sh override 82 "<reason>".

[bryantharpe]

CI repair — authoring seat had already exited

rustfmt and clippy were both red on this PR. The authoring polecat (furiosa) had already run gt done and exited (checkout back on main, zero commits), so no seat remained to own the red checks. This is the second live occurrence of ps-ops-17-fmt-pre-done-gate, after PR #80.

Repaired by hand in 90254a5. All 12 checks are now green.

Changes were kept deliberately minimal — the authoring design is untouched, so the judge reviews furiosa's work, not mine:

Fix	Location	Nature
cargo fmt --all	refine.rs, solve.rs, verify.rs	whitespace only
clippy::question_mark	verify.rs:77	match { Some(r) => r, None => return None } → ?; semantically identical
clippy::too_many_arguments (8/7)	verify.rs:204	#[allow]
clippy::type_complexity (5-tuple return)	verify.rs:213	#[allow]

On the two #[allow]s: I used suppressions rather than redesigning match_projected_to_centroids, matching existing precedent in this codebase (solve.rs:42, solve.rs:80, preparation.rs:104 all carry #[allow(clippy::too_many_arguments)] from previously-merged PRs). If the judge thinks the 5-tuple return and 8-arg signature deserve a real refactor, that is a legitimate finding and should be filed — I deliberately did not pre-empt that call on someone else's PR.

Verified locally before push: cargo fmt --all --check clean, cargo clippy --workspace --all-targets --all-features --locked -- -D warnings clean, 137 tests pass.

Note for ps-ops-17: clippy escaped here too, not just rustfmt — that widens the bead's defect statement. Detail in the bead comment.

[bryantharpe]

review/judge — APPROVE on 90254a5b2b71

VERDICT: APPROVE
BLOCKING: NONE
NITS: crates/plate-solver/src/verify.rs:267 — .expect("matched catalog star index valid") is a panic vector; consider returning an error or filtering invalid indices gracefully instead.
NITS: crates/plate-solver/src/refine.rs — no unit tests for refine_solution or refine_camera; consider adding tests to lock in the refinement behavior.
NITS: crates/plate-solver/src/refine.rs:30 — _rotation is computed but unused; the refined rotation is not stored in the Solution, only its RA/Dec/Roll decomposition — confirm this is intentional per spec.

---

Independent U3 review by the judge role (glm-5.2, Zhipu — different lineage
than every author seat). Verdict is on blocking findings, not a score. To
supersede with human sign-off (U4): judge-pr.sh override 82 "<reason>".