feat(insights): add new batch worker for event processing

[rabidpraxis]

This adds an EventsWorker, with similar logic as the one added for the Elixir client.

It gathers batches of events and if it either crosses the events_timeout or passes the events_batch_size queue length, it will attempt to send the batch. Each batch gets 3 tries before being dropped and they are independently counted.

If the backend returns a throttling response (429) the events worker will wait for events_throttle_wait before attempting to send a batch payload. Batches will be dropped in FIFO order after events_max_batch_retries attempts. This is also the case with throttling to keep the set of events fresh for when the backend resets its throttle (or the user upgrades)

We drop events after events_max_queue_size has been reached (including the counts of events that are waiting in batches) and periodically log the dropped counts.

On top of adding the events worker, I added some small fixes, rewordings, and QoL things like types and protocols. I added a mypy check to pytest so types are checked as part of the test runs.

[Copilot]

Pull Request Overview

This PR adds a new asynchronous EventsWorker to batch and send events to Honeybadger Insights, and it makes minor refactorings and type improvements throughout the codebase. Key changes include:

• Introduction of the EventsWorker and associated types, configuration options, and connection protocol updates.
• Updates to tests and integrations (Flask, Django, ASGI, AWS Lambda) to support the new events workflow.
• Minor refactors in logging, naming, and type annotations across several modules.

Reviewed Changes

Copilot reviewed 23 out of 23 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
honeybadger/version.py	Updated version string quote style.
honeybadger/types.py	Added events-related types and enums.
honeybadger/tests/*	Updated tests to validate new events handling and add type ignores where needed.
honeybadger/protocols.py	Introduced the Connection protocol including send_events.
honeybadger/fake_connection.py	Renamed send_event to send_events and updated logging messages.
honeybadger/events_worker.py	New file implementing asynchronous batching and sending of events.
honeybadger/core.py	Updated event method to integrate the EventsWorker and use updated timestamp.
honeybadger/contrib/*	Minor adjustments and type ignores to support modern language features.
honeybadger/connection.py	Updated request sender to return structured events send results.
honeybadger/config.py	Added new events_worker configuration options with default values.
dev-requirements.txt	Added additional type checking dependencies.
README.md	Updated documentation to reflect the new events_worker configuration options.

Comments suppressed due to low confidence (1)

honeybadger/config.py:30

• The conversion of boolean options using bool(val) may yield unexpected results because non-empty strings (e.g., 'False') evaluate to True. Consider implementing a more robust boolean converter that explicitly checks for string values 'true' or 'false'.

elif option_types[option] is bool:

[roelbondoc]

This looks good to me!

[subzero10]

I had a run through this, but not in great detail because I have little to none multi-threading experience in python. I'll probably give it another go, but here's some initial feedback.

One issue that we've had in other languages (PHP, Node.js) is that our package may run inside an AWS Lambda function:
The process may exit while there are events in the queue. Therefore we should be able to hook into the shutdown flow and flush the queue if necessary. Do you think we should handle this scenario here as well?

[rabidpraxis]

Thanks for the review @subzero10. I had thought that atexit would deal with TERM signals, but it does not, so I added a signals handler to shutdown (thus cleaning up) the process if something else is not handling it.

This works for the few tests that I ran it through, but as I am finding with many things in the Python instrumentation space, despite PEP 20, there is not one easy way to detect an exiting process :)