Tolerate truncated / partially-corrupt FSSHTTPB streams

[orangeruan128]

Summary

Add local error recovery to four read sites in FileNode.py so a single truncated or partially-corrupt FSSHTTPB structure no longer aborts parsing of the entire document. No behavior change for well-formed input.

Failure modes covered

Each of the four cases below was observed on real-world .one files where parsing currently aborts even though the rest of the document is well-formed:

1. ValueError: cannot fit 'int' into an offset-sized integer

FileNodeList.__init__ calls file.seek(file_chunk_reference.stp). When stp is the result of an unsigned read but the underlying file object treats it as a signed off_t, the seek raises before the body is even read. Fix: catch OverflowError / ValueError / OSError and treat the chunk as an empty list.

2. struct.error: unpack requires a buffer of N bytes — header

ObjectSpaceObjectStreamOfIDs.__init__ reads ObjectSpaceObjectStreamHeader eagerly. On a truncated stream this raises before self.header is ever assigned, so callers checking self.header.OsidStreamNotPresent or self.header.ExtendedStreamsPresent would also crash. Fix: catch struct.error and synthesize an empty header (Count=0, OsidStreamNotPresent=True, ExtendedStreamsPresent=False).

3. struct.error: unpack requires a buffer of 4 bytes — body / 2 bytes — PropertySet

ObjectSpaceObjectStreamOfIDs body loop and PropertySet.__init__'s cProperties read can both run past EOF on truncated streams. Fix: catch struct.error and break / treat the set as empty.

4. KeyError in CompactID.__str__ / __repr__

CompactID.__str__ does document._global_identification_table[current_revision][guidIndex]. Two real-world cases miss the lookup:

• guidIndex == 0xFFFFFF (16777215) — the documented "invalid" sentinel ([MS-ONESTORE] CompactID).
• Cross-revision references whose target table is not yet populated.

Both currently raise KeyError from inside __str__/__repr__, which propagates out of the JSON serialization and aborts the whole document. Fix: factor the lookup into _resolve_guid() and return '<unresolved guidIndex=0xNNNNNN>' on miss, so the rest of the JSON output is still produced.

Scope / non-goals

• Pure error recovery. The happy path (well-formed input) reads the same bytes in the same order and produces the same parse tree.
• No changes outside pyOneNote/FileNode.py.
• Independent of Implements candidate code for pocessing ArrayOfPropertyValues #16, Ensure FileNode.data is initialized in __init__ and is not used if it… #20, and Make property/file content stringification type-safe #26 — each of those targets a different root cause and can be merged separately. With all four merged, parsing tolerance on real-world OneNote backups improves substantially.

Verification

• ast.parse on the modified file passes.
• On a corpus of real-world .one files that previously aborted with each of the four error messages above, parsing now completes and produces the expected JSON output (pages, properties, embedded files). Files that already parsed cleanly continue to produce byte-identical JSON for their headers / properties / files structure.