🐛 fixup request cancellation for v0.6.5

[joerunde]

Compatibility fix for vllm v0.6.5

Description

vLLM has removed the cancellation polling in iterate_with_cancellation and merge_async_iterators, moving it up to the http request handler. This PR removes those integrations, and adds logging integration when requests are cancelled.

It turns out the async grpc server implementation will cancel the task running the handler when the context cancels, so there is no need to poll or implement our own cancellation handling anyway. The LLM engine implementations in vLLM already also catch asyncio.CancelledError and abort the work in the engine for us.

TL;DR: We didn't need this anyway

How Has This Been Tested?

Manual verification of logs and that work is removed from the engine's queues

Merge criteria:

• The commits are squashed in a cohesive manner and have meaningful messages.
• Testing instructions have been added in the PR body (for PRs involving changes that are not immediately obvious).
• The developer has manually tested the changes and verified that the changes work

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 33.33333% with 4 lines in your changes missing coverage. Please review.

Project coverage is 62.98%. Comparing base (04cafa2) to head (20a3586).

Files with missing lines	Patch %	Lines
src/vllm_tgis_adapter/tgis_utils/logs.py	33.33%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #196      +/-   ##
==========================================
- Coverage   63.21%   62.98%   -0.23%     
==========================================
  Files          28       28              
  Lines        1748     1748              
  Branches      216      216              
==========================================
- Hits         1105     1101       -4     
- Misses        537      541       +4     
  Partials      106      106              

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

[fialhocoelho]

LGTM