Support GenerativeModel.count_tokens[_async] for fully-featured generation requests

[adubovik]

It should be possible to count number of prompt tokens via GenerativeModel.count_tokens[_async] method before running an actual generation request.

However, the method takes into account only the contents component of a request.

Other components of a request (such as tools, tool_config, and system_instruction) presumably affect prompt token count, so it would be great to support them in the count-token methods too.

[Ark-kun]

Thank you for the feedback. This makes sense.
However this is currently not supported by the service API:

python-aiplatform/google/cloud/aiplatform_v1beta1/types/prediction_service.py

Line 702 in cd85d8f

class CountTokensRequest(proto.Message):