[Python] Trust dataset indexes and expose bounded read caches

[ColinLeeo]

Indexed dataset reads currently recheck source file generations and retain prepared-series metadata without a bound. Add trust_index=True by default for immutable datasets and expose per-dataset resource limits through TsFileDataFrame.

• trust_index=False preserves source generation checks, stale-index rebuilding at open, and errors when files change during queries. Index format and locator bounds remain validated in both modes.
• Add keyword-only max_prepared_series, descriptor_cache_size, max_open_files, query_workers, and query_parallel_min_rows. Explicit values override environment variables, which override built-in defaults; values are validated before opening resources and fixed for the runtime. Subsets share the same runtime and settings.
• Bound prepared-series and descriptor caches to 4096 entries by default. Zero disables retention. Prepared-series LRU eviction releases idle native handles while pinning active queries and shared aligned-time owners; active queries may temporarily exceed the retention limit.
• Pass trust_index directly through the existing internal C APIs. Preserve the optimized Cython index lookup paths.

use_index=False remains the default. trust_index has no effect in that mode, and explicit resource limits require use_index=True.

Validation on the current develop base:

• C++ library/test binary and Python extensions rebuilt successfully.
• 283 Python tests passed across dataset read options, dataset indexes, and dataset queries, including trusted/strict reads with cache capacities 0, 1, and 2.
• 9 native prepared-series/index tests passed.
• Black, targeted C++ Spotless/clang-format, and git diff --check passed.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.