- Updated nodejs and test harness dependencies
- Re-ran code generators
- Formatted generated code

1.0.64-1 makes @github/copilot a thin loader; the runnable index.js, sdk/,
schemas, and native binary moved into platform packages
(@github/copilot-<platform>-<arch>, exactly one installed). Update all CLI
path resolution to the platform package's index.js (Java uses the loader's
npm-loader.js since Maven properties cannot glob). Also fix downstream type
changes: SQLite params -> unknown, AssistantMessageEventData gains citations,
and Go AttachmentBlob.Data -> *string.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Use Path.Join instead of Path.Combine (never silently drops rooted
arguments) and replace the foreach scan with a LINQ Select +
FirstOrDefault when locating the platform package's index.js.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Format the Node client and update the Go billing test to assert the non-deprecated maxPromptTokens field.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The nodejs-installation fallback in the Java test harness pointed at
@github/copilot/npm-loader.js, which spawns the platform-specific native
binary. The Go, Python, Rust, and .NET harnesses instead glob the installed
@github/copilot-<platform> package and run its index.js under Node.js, the
documented preferred entry point. Switch the Java fallback to the same
copilot-* glob for cross-SDK consistency and resilience.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The bundled CLI 1.0.64-1 gates third-party MCP servers behind the copilot
user's is_mcp_enabled flag: a null/missing value disables them. The e2e
harnesses register a per-token copilot user that omitted is_mcp_enabled, so
the new runtime skipped non-default MCP servers (e.g. meta-echo). That made
the model report the MCP tool as nonexistent, so the next /chat/completions
request no longer matched a recorded snapshot and the replay proxy returned
HTTP 500 (surfacing as CAPIError: 500 Proxy error), failing every MCP e2e
test that exercises a third-party server.

Default is_mcp_enabled to true when serving /copilot_internal/user for a
registered token (explicit values still win), mirroring the connect-proxy
mock. One shared change fixes the e2e harness for all language SDKs.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The 1.0.64-1 CLI gates MCP enablement on a GitHub API call (the MCP registry policy check via /copilot/mcp_registry, plus the /copilot_internal/user lookup that carries copilot_plan and is_mcp_enabled). The Node/Go/Python/.NET/Java e2e harnesses only redirected COPILOT_API_URL to the replay proxy, so these GitHub API calls escaped to the real api.github.com. On macOS CI runners that egress is slow/unreachable, so MCP policy resolution stalled and every MCP server timed out before reaching connected (macOS-only failures; ubuntu/windows reach api.github.com fast enough to pass).

The Rust e2e harness already sets COPILOT_DEBUG_GITHUB_API_URL to the proxy and passes on macOS with the identical CLI and identical MCP lifecycle tests. Mirror that here so the GitHub API calls are served hermetically by the replay proxy: the CLI then reads copilot_plan=individual_pro, the MCP policy resolver short-circuits to allow-all, and the registry-policy network call is never made.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>