fix(llmobs): add type checking to the chat completions endpoint

[Yun-Kim]

This PR fixes three things:

• An issue in the LLMObs' OpenAI integration which stores tool calls (via the Chat Completions endpoint). Chat completions for tool calls return a list of tool calls, but we had previously assumed only one tool call would be returned.
• How we construct streamed tool chat completions. We were previously checking the first chunk in the response to know to join the tool/function_call chunk fields together, but it appears that the first chunk in a response can actually contain no data at all. We are now constructing the streamed response chunk-by-chunk.
• Add type checking for request messages arg in the chat completions endpoint, as OpenAI allows users to pass in OpenAI ChatMessage class types. We were previously only looking for dictionary arguments, but now we'll correctly extract the message content based on the message type.

No changelog is required as this only affects private beta customers for LLMObs.

Checklist

• Change(s) are motivated and described in the PR description
• Testing strategy is described if automated tests are not included in the PR
• Risks are described (performance impact, potential for breakage, maintainability)
• Change is maintainable (easy to change, telemetry, documentation)
• Library release note guidelines are followed or label changelog/no-changelog is set
• Documentation is included (in-code, generated user docs, public corp docs)
• Backport labels are set (if applicable)
• If this PR changes the public interface, I've notified @DataDog/apm-tees.
• If change touches code that signs or publishes builds or packages, or handles credentials of any kind, I've requested a review from @DataDog/security-design-and-guidance.

Reviewer Checklist

• Title is accurate
• All changes are related to the pull request's stated goal
• Description motivates each change
• Avoids breaking API changes
• Testing strategy adequately addresses listed risks
• Change is maintainable (easy to change, telemetry, documentation)
• Release note makes sense to a user of the library
• Author has acknowledged and discussed the performance implications of this PR as reported in the benchmarks PR comment
• Backport labels are set in a manner that is consistent with the release branch maintenance policy

[datadog-dd-trace-py-rkomorn]

Datadog Report

Branch report: yunkim/llmobs-fix-openai-tool-call
Commit report: 2699ecd
Test service: dd-trace-py

✅ 0 Failed, 726 Passed, 2407 Skipped, 15m 35.17s Total duration (1h 4m 13.63s time saved)

[pr-commenter]

Benchmarks

Benchmark execution time: 2024-03-27 18:13:10

Comparing candidate commit 4f20bdb in PR branch yunkim/llmobs-fix-openai-tool-call with baseline commit a04d66b in branch main.

Found 5 performance improvements and 2 performance regressions! Performance is the same for 194 metrics, 9 unstable metrics.

scenario:flasksimple-appsec-get

• 🟥 execution_time [+214.817µs; +262.118µs] or [+3.433%; +4.189%]

scenario:httppropagationextract-large_header_no_matches

• 🟩 max_rss_usage [-735.659KB; -669.679KB] or [-3.365%; -3.063%]

scenario:httppropagationextract-medium_header_no_matches

• 🟩 max_rss_usage [-770.653KB; -695.715KB] or [-3.523%; -3.181%]

scenario:httppropagationextract-none_propagation_style

• 🟩 max_rss_usage [-1011.664KB; -932.298KB] or [-4.624%; -4.261%]

scenario:httppropagationextract-wsgi_invalid_trace_id_header

• 🟥 max_rss_usage [+531.474KB; +712.891KB] or [+2.514%; +3.373%]

scenario:httppropagationextract-wsgi_large_header_no_matches

• 🟩 max_rss_usage [-797.485KB; -719.674KB] or [-3.639%; -3.284%]

scenario:httppropagationextract-wsgi_medium_header_no_matches

• 🟩 max_rss_usage [-767.967KB; -685.294KB] or [-3.512%; -3.134%]