OMM interface overhaul: full-fidelity dicts, file/text loaders, TLE conversion, one parser for Rust and Python

[ssmichael1]

Follow-up to the OMM interface review. Closes the gaps between the Rust and Python OMM paths and exposes what Rust already parsed.

Python

• omm_from_url is no longer lossy. It returns every field the source provided: version, originator, COMMENT, CENTER_NAME/REF_FRAME/TIME_SYSTEM/MEAN_ELEMENT_THEORY, GM, mass/areas/coefficients, BTERM/AGOM, and all extras (Space-Track's OBJECT_TYPE, RCS_SIZE, LAUNCH_DATE, TLE_LINE1, …; XML USER_DEFINED parameters). Numbers come back as numbers, EPOCH as RFC 3339.
• New omm_from_file(path) and omm_from_text(text), JSON or XML detected from content. The tests and the OMM tutorial no longer need xmltodict (dropped from the test extra and docs/requirements.txt).
• TLE.from_omm(dict) / TLE.to_omm().
• One parser. sgp4 dict inputs are flattened to JSON and handed to the same serde parser as Rust, replacing ~130 lines of hand-rolled field code. Verified behaviours: quoted numbers, None/"" optionals, sgp4 / utc metadata accepted, any level of an xmltodict tree (omm, body, segment, data node), EPOCH as str/satkit.time/datetime, numpy scalars, and an error that names the expected keys instead of a bare "OMM epoch is required".
• SGP4-XP sets are refused. EPHEMERIS_TYPE: 4 raises instead of running classic SGP4 silently.
• OMMDict TypedDict (runtime module satkit/_types.py, re-exported at top level) documents the keys and units; sgp4 is typed TLE | OMMDict | list[TLE | OMMDict]. stubtest passes.
• omm_from_url releases the GIL during the fetch.

Rust (breaking, for 0.22)

• OMM.epoch: Instant (was String; epoch_instant() kept as deprecated). No reparse per call, no Instant::INVALID fallback.
• OMM derives Serialize (absent optionals skipped, EPOCH written as RFC 3339, extras written back).
• Default removed; OMM::from_mean_elements(epoch, n, e, i, Ω, ω, M) replaces it.
• OMM::from_tle / OMM::to_tle plus From<&TLE> / From<&OMM> impls; OBJECT_ID ↔ international designator conversion.
• from_json_value, from_text, from_file, reset_cache; from_json_string accepts a bare object (the from_url string hack is gone).
• One generic tolerant scalar parser replaces the three de_opt_* and four parse_optional_* helpers.
• XML: COMMENT lines (header, metadata, data), CREATION_DATE, spacecraftParameters, and userDefinedParameters are read, so Space-Track XML and JSON now yield identical messages. OBJECT_NAME/OBJECT_ID default to UNKNOWN (trimmed dicts).
• omm::Error is #[non_exhaustive]; InvalidFloat/InvalidInt/InvalidFloatField collapse into InvalidField { field, message }; new UnsupportedEphemerisType, UnexpectedJsonShape, UnrecognizedFormat.
• Module docs rewritten; leftover "remove this branch if…" comments gone.

Not done: KVN parsing, covariance block, and a Python OMM class (dicts stay the Python representation on purpose).

Tests

• Space-Track ships TLE_LINE1/2 with each OMM, so both parsers are checked against each other over the corpus (Rust: 200 records, Python: 100) — within 10 m at epoch + 3 h (the TLE epoch's 0.864 ms resolution bounds the difference; worst case seen 2 m).
• Space-Track XML vs JSON parity (all 2000 records load; first record field-by-field identical).
• Serialize round-trip, TLE round-trip (to_2line identical), tolerant scalars, rejection cases, cache-reset behaviour, xmltodict shapes, EPOCH variants.
• cargo test --lib 305 passed, cargo test --doc omm 4 passed, pytest python/test/ 197 passed, stubtest clean, clippy -D warnings clean, notebook cells executed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GTxc7VmUFyqdpobZkg7Rb7