fix: make Redis checkpoint listing conformant

[CreativeSystemsDevelopment]

Summary

• filter empty checkpoint namespaces explicitly instead of treating them as absent
• allow custom metadata filters by post-filtering stored checkpoint metadata
• apply before pagination consistently for checkpoint IDs that are not ULIDs
• make prune([]) a no-op for async/sync Redis savers
• add sync and async regression coverage for namespace, custom metadata, and before pagination

Validation

• PYTHONPATH=/tmp/langgraph-redis-upstream agent_server/.venv/bin/python -m py_compile langgraph/checkpoint/redis/base.py langgraph/checkpoint/redis/__init__.py langgraph/checkpoint/redis/aio.py tests/test_checkpoint_filters.py
• LangGraph langgraph-checkpoint-conformance==0.0.2 against native Redis 8.6.2: FULL, all detected capabilities passed
• Direct sync/async regression script against native Redis 8.6.2: root namespace, custom metadata, and before pagination passed

Docker/Testcontainers pytest was not run locally because Docker is not installed in this WSL distro; the repo CI should exercise the containerized fixture.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5494d2e7a1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Skip run_id in metadata post-filter checks

When list/alist falls back to post-filtering (for example filter={"run_id": "r1", "score": 42}), _metadata_filter_matches re-validates filter keys against metadata but only skips thread_id. run_id is indexed separately and is often absent from metadata when it comes from config, so valid results are dropped after the Redis query already matched run_id. This makes mixed indexed+custom filters return empty/incomplete results in both sync and async listing paths.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Respect caller limit in post-filter query window

In post-filter mode the search always fetches only the first 10,000 docs, then applies metadata/before filtering in Python. If matching checkpoints are beyond that initial window (or the caller requests limit > 10000), they can never be returned even though they exist, so results become silently incomplete. This affects pagination/filter correctness whenever histories are larger than 10k entries.

Useful? React with 👍 / 👎.