[Feat][Router] Add per-model request latency histogram (#940)

vllm:request_latency_seconds with server/model/status labels. observed
once per request in finally, so http 4xx/5xx and exceptions both land
as status=error.

Refs #699.

Signed-off-by: Mikhail Basov <Michael.S.Sinclair@protonmail.com>
Co-authored-by: Rui Zhang <51696593+ruizhang0101@users.noreply.github.com>