docs: track Custom endpoints per-model routing plan

[Bionic711]

Summary

• Add the planning specification for per-model API suffixes, explicit Custom API type and URL Handling, APIM prefix preservation, and a proposed named GenAI.mil profile.
• Record selective reuse of the pinned microsoft/simplechat Paullizer branch without importing its React UI or unrelated features.
• Defer Responses and custom calls to future plans while retaining bounded routing/adapter extension points. Exclude agent changes and preserve existing behavior.
• Keep this PR in draft for tracking and design review. It contains documentation only, not runtime implementation, migration, deployment, or approval to perform those actions.

Open Decisions

• Confirm whether Azure/Foundry Chat Completions is a separate labeled dropdown choice sharing the OpenAI-compatible adapter, distinct from Azure deployment-style routing.
• Complete review of the proposed GenAI.mil support scope and capability limits. Supplied API documentation was not live-verified.
• Obtain explicit implementation approval after the remaining design review.

Linked Issue

No issue assigned. Existing Custom endpoint work references #1222, but this PR does not claim that issue covers or is resolved by the proposed scope.

Release Notes & Latest Features

• Internal only: planning documentation; no new end-user or admin functionality in this PR.
• No Latest Feature card or screenshot needed for this documentation-only tracking change.

Version Bump

• Application version bump not needed because this is docs-only; baseline remains 0.261.039.
• Deployer version bump not needed because deployers are unchanged.

Testing / Validation

• Passed focused document checks for scope boundaries, active JSON examples, 13 unique acceptance criteria, relative links, and whitespace.
• git diff --cached --check passed before commit.
• Editor diagnostics reported no errors in the specification.
• No runtime regression suite or live inference tests run for this documentation-only change. The spec distinguishes prior network-blocked/mock findings from unverified live behavior.

Documentation

• Feature planning document added: docs/explanation/features/CUSTOM_ENDPOINTS_PER_MODEL_ROUTING_SPEC.md.
• Release notes not changed: no shipped behavior or implementation version yet.
• Fix documentation not needed: this PR does not implement the APIM fix.

Security Checklist

• No Flask routes, frontend settings payloads, or browser assets changed.
• No credentials, keys, connection strings, customer data, or unrelated local artifacts included.
• Planned work retains authorization, secret handling, outbound URL/TLS protections, and fail-closed behavior for deferred API types.