Fixes spans in streaming responses when using WSGI based frameworks. Only close the transaction once the response was consumed. This way all the spans created during creation of the response will be recorded with the transaction:

- The transaction stays open until all the streaming blocks are sent to the client. (because of this I had to update the tests, to make sure the tests, consume the response, because the Werkzeug test client (used by Flask and Django and our Strawberry tests) will not close the WSGI response)
- A maximum runtime of 5 minutes for transactions is enforced. (like Javascript does it)
- When using a generator to generate the streaming response, it uses the correct scopes to have correct parent-child relationship of spans created in the generator.

People having Sentry in a streaming application will:
- See an increase in their transaction duration to up to 5 minutes
- Get the correct span tree for streaming responses generated by a generator

Fixes #3736

The docker image name of the official Clickhouse docker image changed, so I updated our GH action that starts that docker container and reference the new version here.

The latest release of httpx seems to have broken the test clients of some older versions of Litestar, Starlite, Anthropic, Langchain, OpenAI, Starlette.

Pinning httpx for old versions.

Also tweaking what versions to test against.

* build(deps): bump codecov/codecov-action from 5.0.2 to 5.0.7

Bumps [codecov/codecov-action](https://github.com/codecov/codecov-action) from 5.0.2 to 5.0.7.
- [Release notes](https://github.com/codecov/codecov-action/releases)
- [Changelog](https://github.com/codecov/codecov-action/blob/main/CHANGELOG.md)
- [Commits](codecov/codecov-action@v5.0.2...v5.0.7)

---
updated-dependencies:
- dependency-name: codecov/codecov-action
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <[email protected]>

* template

---------

Signed-off-by: dependabot[bot] <[email protected]>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Ivana Kellyer <[email protected]>

* Fix asyncio testing setup

* default `asyncio_default_fixture_loop_scope` to `function` to get rid
  of deprecation messages
* Change `test_asyncio.py` event loop scopes to `module` to avoid that
  event loop bleeding into all other tests in the same `session`.
* Remove explicit `event_loop`s since `pytest-asyncio` takes care of
  those
* Bump asyncio tests to 3.8 min

…3798)" (#3836)

This reverts commit da20623. (PR #3798)

Having a timer thread on each request is too much overhead on high volume servers.

* Make ray tests actually test something and show that actors are not supported

* fix(grpc): Return propagate proper metadata object instead of list in client interceptor

Fixes #2509

* fix(grpc): Transform metadata into Metadata object in case it's a tuple

Up until version 1.65.0 of grpcio, the metadata was not guaranteed to
arrive as the type specified in annotations but could be a tuple.
To support versions before that we check and transform it here.

* docs(grpc): Add comment about workaround

---------

Co-authored-by: Anton Pirker <[email protected]>
Co-authored-by: Daniel Szoke <[email protected]>

Add a new `init()` option `add_full_stack` (default `False`), when set to `True` it will add all the missing frames from the beginning of the execution to the stack trace sent to Sentry. 

Also adds another option `max_stack_frames` (default `100`) to limit the number of frames sent. The limitation is only enforced when `add_full_stack=True` to not change behavior for existing users. 

Fixes #3646

…le (#3425)

A simple script that parses all libraries we test against from our `tox.ini` and then checks PyPI if this library already supports the newest Python version (currently 3.13)

I was testing Spotlight with Sentry and realized things started to get slow and crashy. It looks like sometimes `args` is just an empty array on cache's `_instruments_call` causing lots of exceptions being thrown. This patch fixes that with explicit length checks and also adds a note for the missing instrumentation for `get_or_set` method. This might be related to #2122 and #3300.

Current Spotlight error handling logic gives up sending events to Spotlight after 3 errors. This doesn't make much sense because:

1. Since there is no back off or retry mechanism, even a very brief server hiccup or restart turns off Spotlight reporting
2. Once this shut off kicks in, there is no way to turn it back on except for a server restart

I added a note for future work for retries and some short buffer.