S1 RTC: discover_s1tiling_acquisitions returns 0 for multi-frame products (masked acq timestamp)

[lhoupert]

Summary (F1)

discover_s1tiling_acquisitions silently discovers 0 acquisitions for multi-frame S1Tiling products, because their GeoTIFF filenames carry a masked acquisition time (…txxxxxx) that the filename regex rejects.

Surfaced by the 32TLR staging e2e (data-pipeline tracker #226, item F1). data-pipeline shipped a workaround in #237 (ingest_all rewrites …txxxxxx… names from the GeoTIFF ACQUISITION_DATETIME tag before discovery). This issue tracks the durable upstream fix so that workaround can be removed.

Root cause

src/eopf_geozarr/conversion/s1_ingest.py:

S1TILING_FILENAME_PATTERN = re.compile(
    ...
    r"(?P<acq_stamp>\d{8}t\d{6})_"   # requires 6 time digits
    ...
)

A multi-frame product names the time portion txxxxxx (e.g. s1a_32TLR_vv_DES_037_20260521txxxxxx_GammaNaughtRTC.tif) because the frame spans acquisitions without a single shared timestamp. \d{6} fails to match xxxxxx, so:

1. parse_s1tiling_filename() returns None,
2. discover_s1tiling_acquisitions() skips the file,
3. discovery returns an empty list — no error, just 0 acquisitions.

Reproduce

Point discover_s1tiling_acquisitions(input_dir) at a directory whose GeoTIFFs have a masked …txxxxxx stamp → returns [].

Proposed fix (durable)

The real timestamp is already available in the GeoTIFF metadata — extract_geotiff_metadata() reads the ACQUISITION_DATETIME tag (s1_ingest.py:148). So:

1. Relax acq_stamp to also accept a masked time, e.g. (?P<acq_stamp>\d{8}t(?:\d{6}|x{6})).
2. When the stamp is masked, resolve the real acq_stamp from the GeoTIFF ACQUISITION_DATETIME tag and use that for grouping (so the per-acquisition key + downstream STAC datetime are correct).

This moves the data-pipeline #237 normalization into discovery, where it belongs, and lets that workaround be retired.

Acceptance

• A masked-stamp multi-frame bundle is discovered as a complete acquisition (vv/vh/masks grouped), with acq_stamp resolved from ACQUISITION_DATETIME.
• Unmasked single-frame products are unchanged (regression-tested).
• data-pipeline #237 workaround can be removed once this lands.

References

• data-pipeline #226 (e2e tracker, item F1), #237 (workaround)
• Tracked alongside the S1 RTC builder work (feat/s1-rtc-stac-builder, PR Feat/s1 rtc stac builder #180)