Support speculative decoding with llama.cpp

[kerthcet]

What would you like to be added:

We have supported vllm, since llama.cpp adds this feature, we should support it as well, see ggml-org/llama.cpp#10455

Why is this needed:

Completion requirements:

This enhancement requires the following artifacts:

• Design doc
• API change
• Docs update

The artifacts should be linked in subsequent comments.

[kerthcet]

/assign

[kerthcet]

/kind feature

[kerthcet]

/help

[kerthcet]

/unassign

[cr7258]

/assign