Add dedicated --allow-testing-features gate for builtinTestServer

[MGudgin]

Summary

This PR fixes a fail-open parity gap where the builtin test proxy
(network.proxy.builtinTestServer) could be activated on the Windows
process-container backends (AppContainer and BaseContainer) without any gate,
while bubblewrap gated it behind the overloaded --experimental flag. It
introduces a dedicated testing-only axis, --allow-testing-features, enforced
uniformly across all backends.

The root cause was that --experimental is a single overloaded axis (backend
admission + experimental features + testing-only helpers) — on the Windows
process-container family it is even self-contradictory as a gate, since it also
flips AppContainer to BaseContainer. The fix decouples "not-for-production
testing scaffolding" into its own flag.

Details

• Add ExecutionRequest.testing_features_enabled, wired from a new
  --allow-testing-features flag on wxc-exec, lxc-exec, and mxc-exec-mac.
• Enforce the gate centrally in validate_common (called by ScriptRunner::run
  for every backend): reject network.proxy.builtinTestServer unless the flag
  is set. This is a distinct axis from --experimental ("unstable/new") versus
  "not-for-production testing scaffolding".
• Remove bubblewrap's now-redundant local --experimental gate; the central
  check covers it.
• The SDK forwards --allow-testing-features automatically when a one-shot
  policy sets builtinTestServer, preserving SDK ergonomics while the direct
  CLI/wxc-exec surface stays fail-closed.
• wxc-test-driver auto-passes the flag for configs that use
  builtinTestServer; the bwrap proxy test script is updated.
• Docs updated: bubblewrap backend, examples, schema, policy v1, and SDK type
  JSDoc.

Tests

• cargo fmt --check, cargo check --workspace --all-targets, and
  cargo clippy --workspace --all-targets -- -D warnings all pass.
• Rust unit tests pass, including two new validate_common tests (reject
  without the flag / accept with it) and an updated bwrap test (wxc_common 340
  passed; plus wxc, lxc, appcontainer_common, wxc_test_driver).
• SDK: tsc build clean and npm test 178 passed, including two new
  flag-forwarding tests.
• macOS mxc-exec-mac compiles on Windows but its runtime path was not
  exercised on this host.

Microsoft Reviewers: Open in CodeFlow

[Copilot]

Pull request overview

This PR introduces a dedicated --allow-testing-features CLI flag and a corresponding ExecutionRequest.testing_features_enabled bit to fail-closed gate the testing-only network.proxy.builtinTestServer feature across one-shot runners, replacing the prior ad-hoc/overloaded gating (notably Bubblewrap’s --experimental-based check).

Changes:

• Add --allow-testing-features to native executors and plumb it into ExecutionRequest.testing_features_enabled.
• Enforce network.proxy.builtinTestServer gating centrally in wxc_common::validator::validate_common, and remove Bubblewrap’s redundant local gate.
• Forward the flag automatically from the SDK (and update docs/tests/scripts to reflect the new gating).

Show a summary per file

File	Description
tests/scripts/run_bwrap_network_proxy_test.sh	Updates Bubblewrap proxy test invocation to pass --allow-testing-features.
src/testing/wxc_test_driver/src/main.rs	Auto-adds --allow-testing-features when configs appear to opt into builtinTestServer.
src/core/wxc/src/main.rs	Adds --allow-testing-features CLI flag and plumbs into one-shot ExecutionRequest.
src/core/wxc_common/src/validator.rs	Centralized validation rejects builtinTestServer unless testing features are enabled; adds unit tests.
src/core/wxc_common/src/models.rs	Adds testing_features_enabled to ExecutionRequest with documentation.
src/core/wxc_common/src/config_parser.rs	Defaults testing_features_enabled to false in parsed requests.
src/core/mxc_darwin/src/main.rs	Adds --allow-testing-features flag and plumbs into request.
src/core/lxc/src/main.rs	Adds --allow-testing-features flag and plumbs into request.
src/backends/bubblewrap/common/src/bwrap_runner.rs	Removes backend-local builtinTestServer gating and adjusts runner test accordingly.
sdk/tests/unit/sandbox.test.ts	Adds unit coverage ensuring the SDK forwards --allow-testing-features when needed.
sdk/src/types.ts	Updates type/JSDoc to document the new gate and SDK auto-forwarding.
sdk/src/helper.ts	Implements SDK auto-forwarding of --allow-testing-features when builtinTestServer is set.
docs/schema.md	Documents builtinTestServer as requiring --allow-testing-features.
docs/sandbox-policy/v1/policy.md	Updates policy docs to mention the new gate for builtinTestServer.
docs/examples.md	Expands examples/docs to explain the new testing-only flag and its rationale.
docs/bwrap-support/bubblewrap-backend.md	Updates Bubblewrap docs to reflect the new flag gating.

Copilot's findings

• Files reviewed: 16/16 changed files
• Comments generated: 2

[MGudgin]

Addressed both review comments in f726343:

• wxc-test-driver: narrowed the match from the bare "builtinTestServer" substring to the concrete "builtinTestServer": true JSON pattern (both spacings), so it can't false-positive on the string appearing in process.commandLine and mask fail-closed coverage.
• policy v1 docs: added { localhost: <port> } as a proxy option in both the type definition and the network table to match the schema/SDK wire format.