Closes #1264.

Problem

The OAuth examples did not consistently use the complete MCP endpoint as the resource identifier:

• the standalone snippets configured the OAuth provider/resource as the origin while connecting at /mcp;
• the simple-auth resource server advertised /mcp even when running SSE at /sse; and
• the bundled client removed every /mcp substring before constructing its OAuthClientProvider.

Strict RFC 9728 clients can reject those mismatches before authorization completes.

Changes

• Pass the exact connection URL to both OAuth client examples.
• Use the selected transport endpoint (/mcp or /sse) as the simple-auth resource identifier and listening route.
• Align the standalone server snippet with MCPServer.run()'s actual default address: http://127.0.0.1:8000/mcp.
• Clarify that AuthSettings.resource_server_url is the complete externally visible MCP endpoint, including mount or proxy prefixes.
• Correct the simple-auth discovery commands and metadata examples.

This PR does not infer or append public paths in SDK code. Internal routes cannot reveal ASGI mount paths or reverse-proxy prefixes. The earlier #2662 explored the same docs/example direction; this revision applies that conclusion to current main and adds focused regressions for the current examples.

Validation

• Focused regressions: 4 passed (Streamable HTTP /mcp, SSE /sse, and exact client URL preservation on asyncio and Trio).
• Example and documentation checks: 213 passed.
• Full suite: 5,577 passed, 16 skipped, 1 xfailed; the only failure is the unchanged Windows symlink fixture, which lacks SeCreateSymbolicLinkPrivilege (WinError 1314).
• Ruff lint and format, markdownlint, EOF checks, and uv.lock consistency passed.
• Focused pyright on all changed Python files and new tests: 0 errors. The all-files Windows hook still reports existing POSIX/Win32 typing errors outside this diff.

AI assistance disclosure: I used Codex to inspect #1264 and the prior PR history, implement and test the changes, and reduce the original draft after review. I independently verified the final diff and the validation results above.