fix: save Redis state for apps without a partition key

[sangkyoonnam]

The Redis persisters now save and load state when the application has no partition key. Today an app built without with_identifiers(partition_key=...) fails on its first step with either persister.

ApplicationBuilder leaves partition_key as None, and RedisBasePersister.save and AsyncRedisBasePersister.save put it inside the hset mapping, which redis-py rejects. Without a namespace, redis-py also rejects None as the sorted-set key the reads use. On main, an app with with_state_persister(persister) and no partition key raises redis.exceptions.DataError: Invalid input of type: 'NoneType' on step(), and the same on astep(). Without a namespace, load(None, app_id) and list_app_ids(None) raise the same error; with a namespace they look up <namespace>:None and find nothing. Either way initialize_from can't resume such an app. The SQLite persisters map None to PARTITION_KEY_DEFAULT ("") in save, load and list_app_ids, the Postgres load does the same, and #949 (open) adds it to the Postgres save, so this aligns Redis with them.

Changes

• RedisBasePersister and AsyncRedisBasePersister get PARTITION_KEY_DEFAULT = "", matching the SQLite and Postgres persisters.
• save, load and list_app_ids on both classes map a None partition key to PARTITION_KEY_DEFAULT and annotate it as Optional[str]. With no namespace the partition index becomes the empty-string key and the state hashes :<app_id>:<sequence_id>; with a namespace they become ns: and ns::<app_id>:<sequence_id>. Redis accepts both, and load returns "" as the partition key, the same as SQLite.

How I tested this

• Three new tests save with partition_key=None, load it back with None, and check that list_app_ids(None) lists the app: the sync persister without a namespace, the sync persister with one, and the async persister. The first one also checks that the row reads back with partition_key == "" and through load("", app_id). All three fail on main with the DataError and pass here.
• The Redis tests ran against a local redis:7 container with BURR_CI_INTEGRATION_TESTS=true: 17 passed (the 14 on main plus the three new ones).
• An app with initialize_from(persister, ...), with_state_persister(persister) and no partition key steps twice, rebuilds from the persister at sequence 1 with the saved state, and steps again, on both persisters.
• pytest tests --ignore=tests/integrations/persisters --ignore=tests/integrations/test_bip0042_bedrock.py: 656 passed, 2 skipped.
• pre-commit run on the changed files passes.

Notes

• load(partition_key, app_id=None) still raises the same DataError. The per-partition sorted set is scored by sequence id, so "latest across app ids" would need a choice the SQL persisters make by created_at, and ApplicationBuilder always sets an app_id, so no app path reaches it. I left it alone.
• Existing data isn't affected. A None partition key never got written before. Apps that passed partition_key="" explicitly now share that key space with apps that passed nothing, which is how the SQLite persisters already behave.
• The key layout joins partition_key, app_id and sequence_id with : without escaping, so an app_id containing : can collide across partitions. That predates this change and needs a migration story, so it stays out of scope.
• feat: durable execution (suspend/resume + ctx.durable journal) #786 adds _partition_key_safe() with "__none__" for the new durable methods on the same classes. If that lands first, one of the two defaults should win. I'd keep PARTITION_KEY_DEFAULT, since SQLite and Postgres already use it.
• The existing Redis tests always pass a partition key, which is why this didn't show up in CI.
• I wrote this with Claude Code and reviewed and ran it myself.

Checklist

• PR has an informative and human-readable title (this will be pulled into the release notes)
• Changes are limited to a single goal (no scope creep)
• Code passed the pre-commit check & code is left cleaner/nicer than when first encountered.
• Any change in functionality is tested
• New functions are documented (with a description, list of inputs, and expected output)
• Placeholder code is flagged / future TODOs are captured in comments
• Project documentation has been updated if adding/changing functionality.

[elijahbenizzy]

Consistent with SQLite and Postgres: the empty string is the persister sentinel, none stays tracking-side. #786 should adopt PARTITION_KEY_DEFAULT rather than the reverse. Elijah