[python] Keep native planning for Blob descriptor reads

[XiaoHongbo-Hope]

Purpose

Keep native scan planning enabled for dynamic reader options which do not affect split planning.

query.read_blobs() copies the table with blob-as-descriptor=true. The native capability gate treated this reader-only option as unsupported and silently fell back to Python planning.

Changes

• allow blob-as-descriptor, read.batch-size, and read.parallelism without disabling native planning; these options remain effective in the PyPaimon reader and are not needed by the Rust planner
• continue forwarding only the scan options explicitly supported by the Rust planner
• keep unknown dynamic options on the conservative Python fallback path because Rust 0.4 does not strictly reject every unknown option
• preserve conservative handling for option removal
• cover Blob descriptor planning and decoding, reader-only options, and unknown-option fallback

Tests

• native plan unit and integration tests: 52 passed, 12 subtests passed
• multimodal read_blobs / stream_blobs tests: 6 passed
• flake8 on changed files
• compileall
• git diff --check
• OSS-backed Data Evolution BLOB table: the high-level query.read_blobs() path invoked native planning once, planned one split at the pinned snapshot, and returned the expected row and Blob payload

[JingsongLi]

+1