[Feature]: Use FP8_e5m2 automatically when using quantized kv cache FP8 on trillium

[bvrockwell]

🚀 The feature, motivation and pitch

Good first issue: Use FP8_e5m2 automatically when using quantized kv cache FP8 on trillium

v6e performs magnitudes better with FP8_e5m2 KV cache quantization as opposed to when --kv-cache-dtype=FP8

Make OOTB performance easier for users by handling this under the hood.

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[zixi-qi]

Created PR to implement this feature: Use FP8_e5m2 automatically when using quantized kv cache FP8 on trillium #1136

Though while testing on v6e8 e2e with vllm bench throughput I ran into an existing kernel issue for fp8 kv, appreciate if I could get help from TPU experts here