test(metadata): Cover metadata exclusion rules in split_metadata()

[shivamm-gupta]

Description

In src/dynavec/metadata.py, split_metadata() partitions a document's metadata between S3 Vectors (small filterable subset) and DynamoDB (full rich metadata).

DynavecConfig provides non_filterable_keys to explicitly exclude certain fields from being sent to S3 Vectors even if they are scalar values. While split_metadata() implements this logic:

for key in candidate_keys:
    if key in config.non_filterable_keys:
        continue
    if key in metadata and _is_s3_filterable_value(metadata[key]):
        s3_meta[key] = metadata[key]

tests/test_metadata.py currently only tests the filterable_keys allowlist (test_split_respects_filterable_keys_allowlist) and does not test non_filterable_keys or its precedence rules.

Proposed Scope & Solution

Extend tests/test_metadata.py with comprehensive unit tests for non_filterable_keys:

1. Direct exclusion: An excluded key present in non_filterable_keys stays in the DynamoDB metadata copy, but is absent from the S3 Vectors filter metadata.
2. Precedence over allowlist: When a key is present in both filterable_keys (allowlist) and non_filterable_keys (denylist), exclusion in non_filterable_keys takes precedence.
3. Namespace isolation tag: Confirm that non_filterable_keys cannot inadvertently strip the mandatory namespace tag (__ns).
4. Immutability of input: Verify that the caller's input metadata dictionary is not mutated during the split.

Acceptance Criteria

• Unit tests added to tests/test_metadata.py covering all the scenarios above.
• Tests run 100% offline without live AWS credentials.
• Ruff / lint passes clean.

[GITHUBTANMAY2003]

Hi, I’ve reviewed the issue, the relevant split_metadata() implementation, and the existing tests. I understand the required changes and would like to work on this issue. Could you please assign it to me?