Add RecordingRunner and Execute() test suite

[kindermax]

Parent

#355

What to build

Create a RecordingRunner test helper (satisfies the ScriptRunner seam from #362) that captures every script invocation — the script text, the env slice, and the working directory — and can be configured to return controlled errors on specific calls.

Using RecordingRunner, write a comprehensive table-driven test suite against Executor.Execute(). The test file lives in package executor (same package, internal access) and uses only the standard testing library — no third-party assertion frameworks.

Tests to cover:

• Init script runs on the first Execute() call and is skipped on subsequent calls
• Dependencies execute in declaration order
• A dependency failure returns a DependencyError with the correct command chain
• Sequential command scripts execute in order
• after script is invoked even when the main command script fails
• after script error does not override the main command's exit code
• Parallel mode dispatches all command scripts (order-independent)
• A failure in one parallel script causes Execute() to return an error
• LETS_CHECKSUM_* env vars are present in the runner invocation when a checksum is defined
• Checksum is persisted after a successful execution
• Checksum is not persisted when a command script fails

Acceptance criteria

• RecordingRunner captures script, env, and dir per invocation and supports returning controlled errors
• Test covers: init runs once across multiple Execute() calls
• Test covers: depends run in declaration order
• Test covers: dependency failure produces correct DependencyError chain
• Test covers: sequential scripts run in order
• Test covers: after-script runs on main-script failure
• Test covers: after-script error does not change the returned exit code
• Test covers: parallel mode dispatches all scripts
• Test covers: parallel failure propagates as an error
• Test covers: checksum env vars present in runner invocation
• Test covers: checksum persisted only on success (not on failure)
• All tests use package executor (internal), standard testing library, no third-party frameworks
• go test ./internal/executor/... passes

Blocked by

• Introduce ScriptRunner seam into executor #362 (ScriptRunner seam must exist before RecordingRunner can satisfy it)