ollama: add tools support

[treywelsh]

In this PR I tried to:

• basically integrate Ollama tools supports in langchaingo
• add an ollama test
• update the ollama-functions-example code now heavily inspired of openai-function-call-example

Notes:

• There's no tool call id management in ollama for now: Support tool/tool call ids when multiple tool calls are requested. ollama/ollama#6708
• some structs are inspired of ollama structs: https://github.com/ollama/ollama/blob/main/api/types.go#L144
  except for Parameters field where I set the any type https://github.com/ollama/ollama/blob/main/api/types.go#L168 for easier integration
• I'm only able to try with smaller models like with llama3.1
• there's potentially a small breaking change, I removed the "omitempty" on the Stream struct field to force specifying the streaming option in the payload even if it's false

The model properly answer with json tool call format but after at end it seems that llama3.1 don't properly use message history with tool results and I don't really know why. It would be interesting to test with other models.

May close #965

PR Checklist

• Read the Contributing documentation.
• Read the Code of conduct documentation.
• Name your Pull Request title clearly, concisely, and prefixed with the name of the primarily affected package you changed according to Good commit messages (such as memory: add interfaces for X, Y or util: add whizzbang helpers).
• Check that there isn't already a PR that solves the problem the same way to avoid creating a duplicate.
• Provide a description in this PR that addresses what the PR is solving, or reference the issue that it solves (e.g. Fixes #123).
• Describes the source of new concepts.
• References existing implementations as appropriate.
• Contains test coverage for new functions.
• Passes all golangci-lint checks.

[tmc]

Lovely, can you get the lint issue addressed?

  Error: llms/ollama/ollamallm.go:49:1: cognitive complexity 36 of func `(*LLM).GenerateContent` is high (> 30) (gocognit)
  func (o *LLM) GenerateContent(ctx context.Context, messages []llms.MessageContent, options ...llms.CallOption) (*llms.ContentResponse, error) { // nolint: lll, cyclop, funlen

[treywelsh]

Fixed.

By the way the ollamaclient in langchaingo is no more up to date (deprecated options, old single embedding endpoint), I may propose a PR soon to remove the ollama internal package from langchaingo and replace it by: https://pkg.go.dev/github.com/ollama/ollama/api

Side note: seems to conflict with at least one other existing example PR: #951
EDIT: And conflict with my new draft PR #1036
This last one should probably be merged first

[pdxrlj]

Is it ready for use now

[atljoseph]

Just bumping this :) Thanks for your contributions.

RE: llama3.1 don't properly use message history with tool results and I don't really know why

• Tool call IDs might help the LLM connect the dots more.

Curious. What style of tool call orchestration is supported? Can the LLM invoke more than one tool call in a response? Can it dictate that specific tools be called concurrently for any reason, or is everything synchronous?

[sami-sweng]

so after a few tests, it looks to me that with this PR the model can correctly see the tools, can correctly query them, but do not see the tool call responses. I might not be using it correctly though.