Record: the decision record matches the pipeline after #897

[cailmdaley]

Brings astra.yaml in line with what the committed pipeline does since #897. Found by a sentence-by-sentence audit of every decision against the code: one cheap reader per decision, then one reviewer per sub-analysis holding the bar "would a reader otherwise believe something false about the catalogue?".

• Detection: the default_tile.sex / config_tile_Sx.ini settings are scoped to the SExtractor (image-sims) path; on data, FLAGS and MAG_AUTO are DR6's. The .sex selectors (DOT_SEX_FILE) are now pinned in Values, so a runner pointed at another file fails the check. detection_threshold_policy now cites the Check that sim-side tile SExtractor reproduces the DR6 detections data uses #929 measurement: on the DR6 image of 186.307 this config reproduces DR6 detection (99.89% complete, 100% pure).
• Masking: says which masks reach pixels. Neighbour segmentation footprints are masked, not only the instrument flags.
• Other decisions: astrometry and epoch provenance name both HISTORY parsers and the ngmix Jacobian; the MCCD refusal at parse time is recorded; the postage-stamp, zero-point, tile-overlap (sp_validation's per-tile box cut) and completeness rationales describe committed data.
• Tags, comments and docs only: CCD_SIZE comments (four strict bounds, not Nx,Ny), the merge_final_cat.py docstring on never-fit rows (it had the flag semantics backwards since Metacal flags are nonzero whenever a type has no measured shear, using ngmix's own bits #854), the TILE_ID comment, and the README mccd lines. No behaviour change.

Unit and module tests in the dev container: 801 passed, 1 skipped. astra-tools validate passes. Pixel-scale and star_selection_box edits ride #927; object_position_columns rides the windowed-positions PR.

Claude Opus 5.5 on behalf of Cail

🤖 Generated with Claude Code

[cailmdaley]

comments/astra only, fine to merge