Add the temporal-covering projection generator

[estebanzimanyi]

Stacks on #24 (the temporal-covering descriptor). Until #24 merges, this PR's diff shows both commits; review/merge #24 first.

Projects the catalog's temporalCovering block (added by #24) onto the canonical, language-agnostic covering-column contract — per temporal type, the ordered covering columns with the fully-composed MEOS expression that derives each from the value. This is the reference rendering every binding generator (PyMEOS, JMEOS, MobilityDuck, MobilitySpark, …) reproduces in its own idiom, so a temporal table prunes the same way on every platform (Iceberg manifest + Parquet row-group min/max).

What it adds

• generator/covering.py — build_covering_projection(catalog), pure dict → dict. Composes each covering column's expression: a box column → accessor(box_from(VALUE)), a value column → accessor(VALUE). VALUE is the temporal column the binding substitutes.
• generate_covering.py — CLI mirroring generate_movfeat_openapi.py; reads output/meos-idl.json, writes output/meos-covering-projection.json.
• tests/test_covering_projection.py — unit + live gate.

Mirrors the existing projection idiom (generator/movfeat.py + root CLI + test).

Projected output (sample)

tgeompoint (STBOX)
  xmin double      stbox_xmin(tspatial_to_stbox(VALUE))
  ...  zmin/zmax [hasZ]
  tmin timestamptz stbox_tmin(tspatial_to_stbox(VALUE))
  srid int         tspatial_srid(VALUE)
tfloat (TBOX)
  vmin double      tbox_xmin(tnumber_to_tbox(VALUE))
  tmin timestamptz tbox_tmin(tnumber_to_tbox(VALUE))

Verification

• 5 projection tests green; live gate against the real catalog: all 11 covered types project to well-formed, value-composed expressions.
• Python-only (no compiled artifact); the build/test is the test suite, green this turn.