cherry-pick: [fix: nvbugs/5355493] Correctly clamp max sequence len to max attention window#5874

MartinMarciniszyn merged 2 commits intoNVIDIA:release/0.21NVIDIA/TensorRT-LLM:release/0.21from

netanel-haber:user/nhaber/cherry-pick/fix/correctly-clamp-max-sequence-len-to-max-attention-windownetanel-haber/TensorRT-LLM:user/nhaber/cherry-pick/fix/correctly-clamp-max-sequence-len-to-max-attention-windowCopy head branch name to clipboard