Skip the state store existence check when setting actor state

[CasperGN]

Based on #1238 (the second PR in this stack, based on #1237 and #1227). Merge it after that. Until then this diff also shows the earlier commits; this PR's own commit is 11f6f53 ("perf(actor): skip the state store read when setting an untracked key").

Description

What was wrong. When set_state or set_state_ttl got a key that was not in the state tracker, it read the state store only to decide whether to record the change as "add" or "update". The state provider saves both as the same upsert, so that read did nothing useful and cost a round trip per new key.

What changed. The read is gone. An untracked key is now recorded as "update" with no store call.

Design choices.

• Update, not add. We no longer know whether the key is already in the store. If the key is set and then removed in the same turn, an "update" becomes a delete, while an "add" is just dropped from the tracker. Dropping it would silently skip deleting a key that does exist. Sending a delete for a key that turns out not to exist does no harm. The .NET change makes the same choice.
• Keys the manager knows are absent still use "add", so add-then-remove still sends nothing. That covers try_add_state after its read, and a cached "not found" entry from the previous PR that is then set.
• try_add_state still reads the store and returns False when the key exists. try_remove_state still reads the store to report whether anything was removed. contains_state is unchanged.
• The default-tracker refresh after a reentrant save only looks at remove versus not-remove, so a reentrant set is saved as an upsert and refreshes the default entry, and a reentrant set followed by a remove sends a delete and evicts the default entry.

Behaviour change. Setting a key that is neither tracked nor known to be absent, then removing it in the same turn, now sends a delete to the store. Before, it sent a delete only if the store read had found the key. The delete still carries the value and ttl from the set (this was already true for set-then-remove on existing keys); the store ignores them, and I left that alone to keep the scope small.

Tests. Existing set_state / set_state_ttl tests now expect "update", and test_get_state_names asserts no store read. New tests cover: set and set_ttl with zero store reads and the exact upserts saved; set-then-remove sending deletes; add-then-remove sending nothing; a reentrant set-then-remove sending a delete and evicting the default entry; and try_add_state / try_remove_state still checking the store. Reverting the change fails 9 tests; keeping the skipped read but recording "add" fails 10, including the delete tests.

This ports dapr/dotnet-sdk#1914. Thanks to @olitomlinson for pointing out the .NET changes on #1227.

Issue reference

Related to dapr/dapr#10532 (reminders miss state saved by reentrant calls) and #1227.

Checklist

• Code compiles correctly
• Created/updated tests
• Extended the documentation (no public API change)

Ran: pytest tests/actor (210 passed), ruff check, ruff format --check, mypy (all clean).

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 84.03%. Comparing base (fb229bc) to head (11f6f53).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1239      +/-   ##
==========================================
+ Coverage   83.89%   84.03%   +0.13%     
==========================================
  Files         123      123              
  Lines       10265    10298      +33     
==========================================
+ Hits         8612     8654      +42     
+ Misses       1653     1644       -9     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.