[codex] add OpenSandbox execution runtime

[jonaswre]

Summary

Adds a repo-authored OpenSandbox execution runtime for Pilot while preserving the existing host runtime behavior.

This PR includes:

• .pilot/environment.yaml parsing and validation
• global runtime config with host and opensandbox providers
• OpenSandbox lifecycle and Execd command client
• sandbox orchestration for clone, checkout, setup, verify, and cleanup
• structured CLI command specs for Claude Code and Qwen Code
• runner wiring for OpenSandbox execution, remote git finalization, branch push, and PR/MR creation
• pilot doctor runtime checks
• runtime examples and configuration docs

Validation

Targeted feature checks pass:

go test ./internal/runtime ./internal/config ./internal/health ./internal/executor -run 'TestEnvironment|TestRuntime|TestConfigRuntime|TestOpenSandboxClient|TestCheckConfig_Runtime|TestRunnerOpenSandboxRuntime|Test.*CommandSpec' -count=1

Also ran:

go test ./...

The full suite still fails in pre-existing baseline areas:

• desktop git graph tests are commit-history-sensitive.
• internal/executor TestAnthropicBackend_IsAvailable depends on local API-key state.