docs(proxy): cover long inference timeouts

[localai-org-maint-bot]

Fixes #11046.

Document the reverse-proxy settings needed by long-running inference and multimodal requests:

• add a valid Nginx location example with read/send/connect timeouts
• raise the request body limit and document buffering controls
• explain why an edge-generated HTML 504 can occur while LocalAI is still running
• distinguish proxy timeouts from the optional LocalAI busy watchdog and its 5-minute default
• mention direct private-network access for bulk jobs

Verification:

• confirmed LOCALAI_WATCHDOG_BUSY_TIMEOUT and its 5m default against core/cli/run.go
• git diff --check
• repository pre-commit hook (docs-only path; code coverage checks correctly skipped)

Hugo and Nginx executables are not installed on this host, so syntax/render validation is delegated to CI.

[localai-org-maint-bot]

@mudler DCO is ACTION_REQUIRED here as well. The commit is bot-authored and repository policy forbids me from adding a human Signed-off-by; please provide maintainer-side attestation if you want this merged. All other CI checks are still running.