Close test-coverage gaps in negative/edge paths (decode_once, CRC/parity failure, health, signal loss)

[zsoerenm]

The v2.0.0 work (#51) has strong positive-path coverage with byte-for-byte golden fixtures (incl. the Spirent L1C-D recording). The systematic gap is in negative and edge paths — several are exercised for GPS L1C-D but not for the other signals, and one keyword is untested everywhere. None of these is a known bug; they're untested code paths that could regress silently.

Gaps

• decode_once=true is never tested (any signal). The early-stop branch in decode (src/gnss.jl, the !decode_once || !is_decoding_completed_for_positioning(...) gate) and every signal's is_decoding_completed_for_positioning are uncovered. Add a test that decodes with decode_once=true and asserts decoding stops once positioning data is validated.

• CRC/parity-failure rejection only tested for L1C-D.

  • Galileo E1B: the crc24q(...) == 0 gate that accepts/drops a word is only ever hit on valid data. Add a corrupted-frame test asserting bad words are dropped.
  • GPS L1 C/A: check_gpsl1_parity and the parity_ok subframe gate are never exercised with bad parity. Add an analogue of the L1C-D corrupt=true test.

• Unhealthy-satellite path only tested for L1C-D. is_sat_healthy is asserted true for Galileo and GPS L1 C/A but never false. The Galileo signal_out_of_service / navigation_data_invalid branch and the GPS L1 C/A svhealth-driven unhealthy branch (incl. the @warn "Bad LNAV Data, SV-Health critical") are unreached. Add unhealthy fixtures for both.

• Mid-stream signal-loss / reacquisition recovery only tested for L1C-D. L1C-D has an excellent reset+recovery regression test; Galileo E1B and GPS L1 C/A have no equivalent for behaviour across a sync discontinuity / outage.

• (minor) decode argument guard untested. decode throws ArgumentError("num_symbols exceeds length(soft_symbols)"); add a @test_throws.

Notes

• The confirm_data voting tests already assert on the returned state's cache, so they're a good template for state-threading assertions.
• Found during review of GNSSDecoder v2.0.0: soft-symbol decoder API + GPS L1C-D decoder #51. Tracking separately so it doesn't block the v2.0.0 release.