Add ability to kill CLI process from test server to test crash recovery scenarios

[varundeepsaini]

Add a KillCaller configuration option to acceptance test server stubs that allows the test server to send a signal (e.g., SIGKILL, SIGTERM, SIGINT) to the CLI process

Motivation
Currently, acceptance tests can verify CLI behavior for successful and error responses, but there's no way to test what happens when the CLI process is forcefully terminated during execution. This will be needed for testing #4090

[varundeepsaini]

Proposed Solution

New header: CLI sends Test-Pid: <pid> on all HTTP requests when DATABRICKS_CLI_TEST_PID=1 is set.
New server stub field: KillCaller (int) specifies the signal to send to the caller process.
Behavior: When a request matches a stub with KillCaller != 0, the server calls syscall.Kill(pid, signal) instead of returning a response.

@denik @andrewnester @pietern @shreyas-goenka @kajaerzen
Could you please review this plan and let me know if more details are needed?
(Also, let me know if I should tag everyone for reviews or not.)
Thankss

[denik]

Sounds reasonable overall.

New header: CLI sends Test-Pid: on all HTTP requests
It might be easier to add something to user-agent, see

bundle/statemgmt/state_pull.go: "github.com/databricks/databricks-sdk-go/useragent"
bundle/statemgmt/state_pull.go: ctx = useragent.InContext(ctx, "engine", string(winner.Engine))

New server stub field: KillCaller (int) specifies the signal to send to the caller process

This is for [[Server]] section in test.toml, right?

[varundeepsaini]

yes

somewhat like this
can change the name to something else if needed

[[Server]]
Pattern = "GET /api/2.0/test"
KillCaller = 9

Also yeah, user agent seems easier