update api.md to make it consist with code.
https://github.com/ollama/ollama/blob/main/server/download.go#L447

Previously our OpenAI chat completions compat layer assumed that tool
calls and content would never be provided together, but this is not a
correct assumption. Content is only optional when tool calls are
present, but tool calls and content can be provided together

Fixes: #11704

openai: allow for content _and_ tool calls in the same message

Added support for converting both `name` and `tool_call_id` fields,
which different clients might provide. `name` is a legacy field from the
OpenAI completions API. For `tool_call_id` we inspect previous messages
and look for a matching tool call ID and grab its name

Issue: #11704

openai: when converting role=tool messages, propagate the tool name

We were missing passing along thinking if content was nil (as opposed
to empty string)

Also added a test for content not being passed, which was the real cause
of <#11704>, since with the way
`Content` is typed, not passing it and empty string are distinct

openai: always provide reasoning

gpt-oss works best with a context length of at least 8k. However,
for GPUs with limited amount of VRAM, there is a significant
performance hit to this increased context. In these cases, we
switch to the Ollama default of 4k