fix(python/sedonadb): serialize pyogrio reader lifecycles

[jiayuasu]

What

• add an ExternalFormatSpec capability for formats that cannot safely consume multiple file readers concurrently
• disable concurrent file reads for pyogrio and serialize schema inference and the complete reader lifecycle within each physical scan
• drop the Arrow FFI stream before closing its owning Python context and releasing the scan permit
• add regression tests for scan-local serialization and deterministic Python-context cleanup

Why

#1051 serialized pyogrio context entry and exit, but active Arrow streams could still overlap. Subsequent wheel runs continued to abort during multi-file and partitioned pyogrio reads, as tracked in #1072.

An unchanged CPython 3.12 comparison repeatedly ran the two affected tests in fresh processes and reproduced SIGABRT at iteration 72 during test_read_ogr_partitioned: run 32625781579.

This change keeps one scan-local permit from reader creation through consumption and native cleanup. Pyogrio multi-file scans therefore trade file-level parallelism for stability. Other external formats retain concurrent reads by default, and independent physical scans do not share a process-wide lock.

Verification

• cargo fmt --all -- --check
• cargo test -p sedona-datasource -p sedonadb --lib (9 + 11 tests passed)
• cargo clippy -p sedona-datasource -p sedonadb --lib --no-deps -- -D warnings
• local pyogrio and external-read modules (22 tests passed)
• pinned CPython 3.12 stress run:
  • full pyogrio module: 18/18 passed
  • 500 same-process repetitions and 500 fresh-process repetitions
  • 2,002 executions of the two affected tests with no abort, hang, timeout, or test failure
  • run 32628586406
• all five core wheel platforms passed on the combined verification branch: run 32627022432

Closes #1072.

Follow-up to #924 and #1051. Replacing the Python reader with the native GDAL reader remains tracked in #1092.