[Frontend] Reduce frequency of client cancellation checking

[njhill]

The api server logic currently checks for client http connection disconnection for each request between every generated token. We can reduce the cost associated with this by limiting the checks to at most once per second.

With llama-3.2-1B, A100 tp=1, --num-scheduler-steps=8:

stream	multistep-stream	req/sec before this PR	req/sec after this PR
false	n/a	95	95
true	false	92	94
true	true	42	67

python benchmarks/benchmark_serving.py \
    --backend vllm \
    --model meta-llama/Llama-3.2-1B \
    --dataset-name sharegpt \
    --dataset-path ShareGPT_V3_unfiltered_cleaned_split.json \
    --port 9999 \
    --save-result \
    --result-dir results \
    --result-filename test.json \
    --num-prompts 2000 \
    --request-rate 200 \
    --seed 42

[mgoin]

LGTM, just one nit about comment placement