fix: skip corrupt/truncated lines in import_session_to_store

[Otis0408]

Summary

import_session_to_store raises json.JSONDecodeError on a corrupt or truncated transcript line, instead of skipping it like every read path in the package.

_append_jsonl_file_in_batches parses each non-blank line with a bare json.loads(line):

for line in f:
    line = line.rstrip("\n")
    if not line:
        continue
    batch.append(json.loads(line))   # <-- unguarded

A partially-written final line — the normal outcome when the CLI is killed mid-append to a .jsonl — therefore aborts the whole import with JSONDecodeError. The read path tolerates exactly this: _parse_transcript_entries (used by get_session_messages) wraps json.loads in try/except (json.JSONDecodeError, ValueError): continue, so the same file reads fine. And because entries flush in batches, a large session leaves the store partially populated before the exception.

The import_session_to_store docstring documents only ValueError (bad UUID) and FileNotFoundError, so the raised JSONDecodeError is also undocumented.

Fix

Skip corrupt/truncated lines, mirroring _parse_transcript_entries. This aligns the behavior with both the documented Raises contract and the package-wide convention of quietly dropping truncated transcript lines (see test_truncated_final_line_is_dropped_not_raised, test_corrupt_lines_skipped).

Tests

Adds test_skips_corrupt_truncated_line: two valid lines followed by a truncated one; asserts the import completes without raising and the store receives exactly the two valid entries (run with batch_size=1 so the earlier batches would already have flushed on the old code). Verified it fails on main with JSONDecodeError and passes with the fix. Full suite: 1058 passed, 5 skipped; ruff and mypy clean.