ps-dbgen-04-cli: implement tetra3-gen-db CLI and reproducibility fixes

[bryantharpe]

Implements the tetra3-gen-db command-line binary for the database-generation crate and fixes the non-determinism that prevented byte-identical output across runs.

Changes

• Add clap-based tetra3-gen-db binary mirroring the reference CLI:
  • Positional STAR_CATALOG and SAVE_AS
  • --max-fov, --min-fov, --linear-probe, --epoch-proper-motion, --star-max-magnitude
  • Tuning knobs: --verification-stars-per-fov, --lattice-field-oversampling, --patterns-per-lattice-field, --pattern-max-error, --multiscale-step
• Re-export propagate and HIP_TYC_PM_ORIGIN from database_generation.
• Fix BSC5 parser to match upstream tetra3 layout (STARN at offset 8, 32-byte entries with f32 id, f64 RA/Dec, i16 spectral type, i16 mag, f32 proper motions in radians/year).
• Pin ZIP last_modified_time to 2000-01-01 for reproducible .npz archives.
• Sort deduplicated patterns before hash-table insertion to remove HashSet iteration-order non-determinism.
• Update catalog_parsing tests for the corrected BSC5 layout.

Verification

• cargo test -p database-generation --locked passes.
• cargo clippy -p database-generation --all-targets --locked is clean.
• cargo deny check passes.
• Two consecutive CLI runs on reference-solutions/cedar-solve/tests/data/bsc5 produce byte-identical .npz files.

🤖 Generated with Claude Code

[bryantharpe]

review/judge — APPROVE on b50ecd37ea2b

VERDICT: APPROVE
BLOCKING: NONE
NITS: crates/database-generation/src/main.rs:172 — when --epoch-proper-motion none is passed, epoch_proper_motion becomes None and the DatabaseProperties.epoch_proper_motion falls back to pm_origin (e.g. 2000.0), which may misrepresent the database metadata since no propagation was actually performed; consider using a sentinel like 0.0 to signal "no propagation."
NITS: crates/database-generation/src/main.rs:248 — name.starts_with("tyc") and name.starts_with("hip") are redundant with the preceding name.contains("tyc") / name.contains("hip") checks since starts-with implies contains.
NITS: crates/database-generation/src/main.rs:155 — config is constructed with epoch_proper_motion but that field is never read from config afterward; only config.star_max_magnitude is used, so the epoch_proper_motion field in the local GenerationConfig is dead in this function.

---

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 81 "<reason>".

[bryantharpe]

review/judge — APPROVE on 53f5892d247c

VERDICT: APPROVE
BLOCKING: NONE
NITS: crates/database-generation/src/main.rs:173 — config is constructed with epoch_proper_motion that duplicates the local epoch_proper_motion variable; consider using the local directly for properties to avoid two sources of truth for the same value.
NITS: crates/database-generation/src/catalog.rs:83 — f32::from_le_bytes(...) as u32 silently truncates/rounds if the f32 catalog number exceeds 2²⁴; a comment noting the safe range would aid future readers, though BSC5 IDs are well within it.
NITS: crates/database-generation/src/main.rs:248 — catalog_name lowercases stem twice in the match arms (once per arm via to_ascii_lowercase()); a single let stem_lower = stem.to_ascii_lowercase() before the match would be cleaner.

---

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 81 "<reason>".