fix(channels): remove empty channel entries on unsubscribe to prevent memory leak

[lesnik512]

Description

ChannelsPlugin._channels grows without bound when the plugin is used with
arbitrary_channels_allowed=True and dynamic channel names (e.g. one channel per user),
because unsubscribe() removed the subscriber and unsubscribed the backend but never deleted
the now-empty set() nor its key from self._channels. For long-lived applications this is a
memory leak that only a process restart clears.

This PR removes a dynamically created channel from _channels once its last subscriber leaves.
Channels declared in the constructor (channels=[...]) are preserved, since they back the
generated WebSocket route handlers and the startup backend subscription.

Reproduction (before this PR)

import asyncio
from litestar.channels import ChannelsPlugin
from litestar.channels.backends.memory import MemoryChannelsBackend

async def main() -> None:
    plugin = ChannelsPlugin(backend=MemoryChannelsBackend(history=10), arbitrary_channels_allowed=True)
    async with plugin:
        for i in range(1000):
            s = await plugin.subscribe([f"user_{i}"])
            await plugin.unsubscribe(s, [f"user_{i}"])
        assert len(plugin._channels) == 0  # fails: 1000

asyncio.run(main())

Changes

• ChannelsPlugin records constructor-declared channels and, in unsubscribe(), deletes a
  channel key once its subscriber set is empty only for dynamically created channels.
• Added tests: arbitrary channel pruned after last subscriber, declared channel kept, and a
  1000-cycle no-leak check. Existing unsubscribe tests use declared channels and are unaffected.

Notes

RedisChannelsStreamBackend.stream_events() keeps a stream_ids cursor map that exhibits the
same unbounded-growth pattern; happy to address it here in a follow-up commit or a separate PR —
let me know your preference.

Closes

Closes #4867

🤖 Generated with Claude Code

---

📚 Documentation preview 📚: https://litestar-org.github.io/litestar-docs-preview/4868

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.29%. Comparing base (64aa847) to head (6fd0f41).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4868   +/-   ##
=======================================
  Coverage   67.28%   67.29%           
=======================================
  Files         293      293           
  Lines       15226    15229    +3     
  Branches     1727     1728    +1     
=======================================
+ Hits        10245    10248    +3     
  Misses       4834     4834           
  Partials      147      147           

☔ 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.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[Vitaly312]

Thanks! The code is fine, just a little nit: comments look a bit verbose, I think we can trim them since the code isn't complicated enough. What do you think?

[lesnik512]

Thanks! The code is fine, just a little nit: comments look a bit verbose, I think we can trim them since the code isn't complicated enough. What do you think?

I've compacted and removed some comments, please take a loog

[Vitaly312]

Thanks. CI failure is unrelated to the changes(looks like some flaky test)

[Vitaly312]

Approved (again?). Anyway, I don't have write access :)