refactor: consolidate OpenAI-compatible provider chat into Nous.Provider macro

[nyo16]

Summary

Six providers — vLLM, SGLang, LM Studio, Mistral, OpenAICompatible, Custom —
carried near-identical chat/2 + chat_stream/2 implementations. The local trio
(vLLM/SGLang/LM Studio) were byte-for-byte structurally identical except for an
id, base URL, env prefix, and display name (a clear three-strikes duplication).

This extends use Nous.Provider with an opt-in chat: config that injects the
shared chat/2, chat_stream/2, base-URL resolver, and header helpers. The
duplication reduces to three variation axes:

Option	Values	Used by
base_url:	:plain · :local · :required	hosted · local servers · user-supplied URL
headers:	:bearer · :bearer_org	with/without openai-organization
timeout: / stream_timeout:	ints	replaces duplicated @default_timeout/@streaming_timeout

Each provider collapses to a moduledoc + a single use block. Adding a new
OpenAI-compatible provider is now just the chat: option.

Changes

• Nous.Provider macro: new chat_ast/1 injects overridable chat/2,
  chat_stream/2, chat_resolve_base_url/1, and chat_build_headers/2.
• vLLM / SGLang / LM Studio / Mistral / OpenAICompatible / Custom: removed
  duplicated bodies, replaced with chat: config.

7 files changed, 233 insertions(+), 329 deletions(-) — ~250 lines of
copy-pasted logic replaced by ~190 lines of reusable macro.

Behavior preservation

This is a behavior-preserving refactor:

• Error message strings reproduced exactly (generated from id/env where
  they were provider-specific — e.g. Custom's SSRF and "requires a base_url"
  messages).
• One deliberate consistency fix: the local trio now forwards finch_name in
  chat_stream, matching the convention already used by the 7 other providers.
  Previously they silently dropped it — they were the outliers.

Verification

• ✅ Provider tests: 197 passed
• ✅ Full suite: 1896 passed, 102 excluded
• ✅ mix credo --strict: 0 issues
• ✅ mix compile --warnings-as-errors: clean
• ✅ mix format --check-formatted: clean