feat: add streaming tool use

[lsorber]

This PR upgrades the chatml-function-calling chat handler with support for streaming tool use and fixes #1883, #1869, and #1756, among other improvements.

Changes:

1. General:
   a. ✨ If no system message is supplied, add an empty system message to hold the tool metadata.
   b. ✨ Add function descriptions to the system message so that tool use is better informed (fixes chatml-function-callling not adding tool description to the prompt. #1869).
   c. ✨ Replace print statements relating to JSON grammars with RuntimeWarning warnings.
   d. ✅ Add tests with fairly broad coverage of the different scenarios.
2. Case "Tool choice by user":
   a. ✨ Add support for more than one function call by making this a special case of "Automatic tool choice" with a single tool (subsumes Support parallel function calls with tool_choice #1503).
3. Case "Automatic tool choice -> respond with a message":
   a. ✨ Use user-defined stop and max_tokens.
   b. 🐛 Replace incorrect use of follow-up grammar with user-defined grammar.
4. Case "Automatic tool choice -> one or more function calls":
   a. ✨ Add support for streaming the function calls (fixes Feature request: add support for streaming tool use #1883).
   b. ✨ Make tool calling more robust by giving the LLM an explicit way to terminate the tool calls by wrapping them in a <function_calls></function_calls> block.
   c. 🐛 Add missing ":" stop token to determine whether to continue with another tool call, which prevented parallel function calling (fixes chatml-function-calling chat format fails to generate multi calls to the same tool #1756).
   d. ✨ Set temperature=0 to determine whether to continue with another tool call, similar to the initial decision on whether to call a tool.

[lsorber]

@abetlen The tests all pass, but the macOS ones were terminated after a timeout. I think this is because of a lack of CPU and or memory resources because the tests run fine on my macOS machine.

[SubatomicPlanets]

I would love to see this merged! Actually there are quite a lot of good pull requests here that i would like to see merged... But this one is top priority!

[lsorber]

Update: I rebased on the latest main and included a few tiny improvements to further improve tool calling robustness.

[lsorber]

Update: I rebased on the latest main and conditionally skipped the added tests on macOS when not enough resources are available to run them.

[LenBanana]

Worked well for me, would you mind rebasing to the latest commit to allow for tool streaming with Qwen models?
Thanks for your work!

[conornash]

Would love to see this merged - is there anything holding it up?

[lsorber]

@abetlen I rebased the PR on the latest upstream main and added a small commit to fix the returned logprobs format.

[tsharp]

I would also like to see this merge. 🧙‍♂️

[XyLearningProgramming]

I rebased this PR onto the latest upstream main (v0.3.16) and resolved the merge conflicts. Opened a new PR at #2129 in case it helps get this merged. All credit to @lsorber for the implementation.