fix: check only decoding requests in _satisfies_last_chunk_constraints

[tjohnson31415]

Description

There was a bug in the scheduler that woud cause no request to be scheduled, putting the scheduler in an infinite loop. This occurred when a very large context-request was being prefilled. When checking _satisfies_last_chunk_constraints, it would say it is unscheduleable becuase it compared the remaining TKV against a copy of itself, exceeding the max_content_len. But it was also the only request running and so nothing would be scheduled.

[github-actions]

👋 Hi! Thank you for contributing to vLLM support on Spyre.
Just a reminder: Make sure that your code passes all the linting checks, otherwise your PR won't be able to be merged. To do so, first install the linting requirements, then run format.sh and commit the changes. This can be done with uv directly:

uv sync --frozen --group lint --active --inexact

Or this can be done with pip:

uv pip compile --group lint > requirements-lint.txt
pip install -r requirements-lint.txt
bash format.sh

Now you are good to go 🚀

[maxdebayser]

Thanks for the fix. As soon as the testing PR is in a working state I'll change the max model len to the value that reproduces this problem before your fix.

[wallashss]

LGTM. Tested and validated.