feat(raster-gdal): add parse_outdb_source helper for the GDAL format driver

[james-willis]

Summary

Adds parse_outdb_source(uri) -> (Cow<str>, u32) in
rust/sedona-raster-gdal/src/source_uri.rs. It is a GDAL-format-driver-internal helper: when an outdb band's outdb_format dispatches to GDAL, the loader calls this to pick the source band out of outdb_uri. Two recognised forms:

• SedonaDB convention (GDAL driver only) — <uri>#band=N → (uri without fragment, N).
  rsplit_once('#') is used so a trailing #band=N wins over an earlier #anchor in the URI.
• GDAL native subdataset URI — HDF5:\"x.h5\":/var, NETCDF:\"file.nc\":var, GTIFF_DIR:N:foo.tif, … → (uri verbatim, 1). We don't try to interpret the GDAL grammar; we just pass it through.
• Plain URI / no fragment / non-band= fragment → (uri, 1).
• Malformed #band= fragment (non-numeric, zero, negative, > u32::MAX) — returns an Execution error. The user explicitly asked for a band; we refuse to silently fall back to band 1.

Returns Cow::Borrowed for both common cases (no allocation when the input has no fragment, and only the trimmed prefix is borrowed when a #band=N is present).

Scope

This is not a SedonaDB-wide URI convention. The schema treats outdb_uri as opaque, and RS_BandPath and other format-agnostic surfaces continue to do so. The #band=N shape only has meaning inside the GDAL format driver — other format drivers (Zarr, MDArray, …) handle their own URIs however they want, and unrelated callers like RS_BandPath are untouched.

Why encode the source band index inside outdb_uri (for GDAL)?

#787 introduced an outdb_band_id: u32 column alongside outdb_url. The follow-up work on N-D rasters needs the GDAL driver to also support GDAL native subdataset URIs (HDF5 groups, multi-page GeoTIFFs, NetCDF variables). Those URIs already carry the source-selector inline, so a typed sibling column would be redundant and couldn't represent them. Inside the GDAL driver, folding both forms (#band=N for plain rasters, native subdataset syntax for everything else) into one parser keeps the driver's URI handling in one place.

Purely additive

No existing call sites are rewired. The first caller will appear in a follow-up PR that ports the GDAL loader to read the source band selector from outdb_uri instead of a separate outdb_band_id column.

Test plan

• cargo test -p sedona-raster-gdal — 43 unit tests + 1 doctest pass
• cargo build -p sedona-schema -p sedona-raster -p sedona-raster-gdal -p sedona-raster-functions -p sedona-testing
• cargo test -p sedona-schema -p sedona-raster -p sedona-raster-gdal -p sedona-raster-functions -p sedona-testing — 315 unit tests + 4 doctests, 0 failures
• cargo clippy -p sedona-schema -p sedona-raster -p sedona-raster-gdal -p sedona-raster-functions -p sedona-testing --all-targets -- -D warnings — clean
• cargo fmt --all --check — clean
• pre-commit run --files rust/sedona-raster-gdal/src/source_uri.rs rust/sedona-raster-gdal/src/lib.rs — clean

The unit tests in source_uri::tests cover: default band 1; explicit band=N; malformed #band= (zero, negative, overflow, non-numeric) returning a clear Execution error; URL query strings preserved before fragment; local paths; HDF5/NETCDF/GTIFF_DIR subdataset passthrough; trailing #band=N winning over earlier #anchor; empty URI; Cow::Borrowed invariants.

[paleolimbot]

A few optional suggestions...thank you!

[paleolimbot]

Optional nit (feel free to merge whenever after an approve when you're happy with it 🙂 )