Use numeric basename as concat index for bare-wildcard rasters (fixes #1465)

[gaoflow]

Problem

When reading externally-provided rasters named with bare numeric stems — e.g. modis_lai stored as 1.tif .. 12.tif and referenced in a catalog with uri: "*.tif" — open_mfraster(..., concat=True) produced the wrong concatenation index and a non-deterministic order (#1465):

• the concat dim was [0, 1, ..., 11] (file-order fallback) instead of the numeric stems [1, 2, ..., 12];
• the order depended on the filesystem glob() order;
• the variable name (and hence the order-dependent result) was non-deterministic.

The index was only derived from a numeric basename when a non-empty prefix had been parsed from the pattern:

elif prefix != "" and bname.split(".")[0].strip(prefix).isdigit():
    index = int(bname.split(".")[0].strip(prefix))
else:
    index = i   # <- bare "*.tif" with numeric names lands here

For a bare wildcard *.tif the prefix is empty, so the numeric stems were ignored.

(This complements #1459, which fixed the write side so hydromt-written mapstacks use <name>_<index>.tif; #1465 is about reading third-party bare-numeric files.)

Fix

• Add an index branch for purely numeric basenames when the pattern has no prefix or postfix (the bare-wildcard case), using the numeric stem as the index.
• Sort the matched files so the concat order and the variable name are deterministic and independent of the filesystem glob order.

Existing prefix / _-separated / PCRaster-style index parsing is untouched.

Reproduction (from #1465)

import tempfile, numpy as np, xarray as xr
from pathlib import Path
from hydromt.readers import open_mfraster

tmp = Path(tempfile.mkdtemp())
for month in range(1, 13):
    da = xr.DataArray(np.zeros((1, 10, 15), "float32"), dims=("band", "y", "x"),
                      coords={"x": np.linspace(0, 7, 15), "y": np.linspace(5, 0, 10), "band": [1]})
    da.rio.write_crs(4326, inplace=True)
    da.rio.to_raster(str(tmp / f"{month}.tif"))

ds = open_mfraster(str(tmp / "*.tif"), concat=True)
print(list(ds["dim0"].values))
# before: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]  (file order, non-deterministic)
# after : [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]  (numeric stems, sorted)

Verification

• New regression test TestOpenMFRaster::test_open_mfraster_bare_wildcard_numeric_names asserts the concat dim equals [1..12]. It fails on main and passes with this change.
• Full tests/data_catalog/drivers/raster/ suite passes (56 passed, 5 skipped).
• ruff check and ruff format --check clean.