[ML 9.3] Add `max_batch_size` parameter for GoogleVertexAI service settings

[kosabogi]

Related dev PR: elastic/elasticsearch#138047

The GoogleVertexAI service_settings object now supports a new parameter: max_batch_size (integer).
This parameter limits the batch size of chunked inference requests sent to GoogleVertexAI.
It can be used together with the max_chunk_size parameter to better control request sizes and avoid exceeding GoogleVertexAI’s token limits.

Required documentation changes

1. Add max_batch_size to the GoogleVertexAI service_settings object.

2. Add a NOTE about the formula estimating a safe batch size:
   Formula:
   batch_size ≈ max_chunk_size * 1.3 (tokens per word) * 512 (max chunks per document) / 20000 (GoogleVertexAI token limit)

3. Update the request example to include this parameter

Additional info: https://docs.google.com/document/d/1ObKSCEJlucp1a3j5iz657aYNqvEvhIoGDRhV5MdYK5Q/edit?tab=t.0