refactor: safe configuration resolution and shared RSM geometry (1/4)

[pecomyint]

Stack 1 of 4. Base main. No beam time required.

Behavior-preserving groundwork. Six-axis Q is bit-identical across the offline, HPC and area-detector paths; the point is to make the next three PRs safe, not to change results.

What this adds

Raw vs effective configuration. IOC_RSM_PARAMETER becomes the canonical geometry section and generates an effective [HKL] mapping in memory. Hand-authored [HKL] on disk is never rewritten and unknown groups survive, so adopting the canonical section is reversible.

Snapshot / CAS persistence. DbProfileConfigSource.save and TomlConfigSource.save both did a shallow existing.update(), so saving one section replaced it wholesale, and the read-modify-write was unguarded — two editors meant last-writer-wins with no warning. Now: load_snapshot() / replace_if_revision() with conflict detection.

One geometry model. RotationAxis / DetectorModel / RSMGeometry and a single build_hxrd, replacing four near-duplicate QConversion + HXRD + init_area sites. Ang2Q.area is called with an explicit deg=True.

Validated SAMPLE_ORIENTATION. Checked against the xrayutilities docstring rather than assumed:

• 'det' requires an innermost detector rotation not parallel to the primary beam, and correctly falls through a final beam-axis rotation (which 'det' ignores) — this is what keeps PR 3's detrot continuous.
• 'sam' warns that it is only physically correct when the innermost sample circle is the azimuth motor. That is a constraint on which motor, not a structural check.
• An orientation collinear with the primary beam is rejected. xrayutilities otherwise silently substitutes a different z-axis — exactly the class of quiet wrong-physics this stack exists to remove. 'det' is the xu default, so this governs every existing profile.

Verification

Full unit suite green, ruff check src/dashpva/ clean, legacy profiles resolve unchanged, and the existing bit-identical-Q test guards the refactor.

Review notes

Merge with a merge commit, not a squash — PRs 2–4 stack on this.

[Osayi-ANL]

Don't see a problem with this commit