fix: prevent server crash on bad function call arguments (closes #64)

[Rohitbhise0004]

Problem

When a MetaCall function is called with the wrong number of arguments (e.g. curl ... -X POST --data '1' against a zero-argument Python function), the native MetaCall runtime throws an exception inside the worker child process. This unhandled error caused the worker process to exit, which:

• Left all pending HTTP requests hanging forever (the HTTP client never received a response)
• • Made the FaaS server appear to be down (metacall-deploy --inspect --dev would exit)
• • Required a full container restart to recover
    Fixes Bug: The local server goes down with bad requests #64.

Root Cause

The proc.on('exit') handler in src/utils/deploy.ts had a // TODO comment acknowledging it could not properly reject in-flight invocations when the worker exited mid-call. Pending invocations stored in invokeQueue were never resolved or rejected, so their HTTP responses were never sent.

Additionally, there was no dedicated error message type for function invocation failures --- the worker reused InvokeResult with a { error: ... } field, which the master process always treated as a success by calling resolve().

Changes

src/worker/protocol.ts

• Add a dedicated InvokeError message type to the protocol enum.

src/worker/index.ts

• In the Invoke handler catch block, send InvokeError (with the stringified error) instead of a fake InvokeResult with an embedded error field.

src/utils/invoke.ts

• Add a pendingIds() method to InvokeQueue so callers can enumerate and drain all in-flight invocations.

src/utils/deploy.ts

• Handle the new InvokeError message type by calling invoke.reject() so the client receives an HTTP 500 with the error message.
• • Fix the exit handler: iterate invokeQueue.pendingIds() and reject every pending invocation --- resolves the // TODO and prevents requests from hanging when the worker crashes.
• • Guard deployReject/deployResolve with an isDeploySettled flag to prevent calling them after the deploy promise has already settled.
• • Add null-guard on invokeQueue.get() return values.

test/test.sh

• Add a regression test in test_python_base_app that POSTs a spurious argument to the zero-arg number() function, asserts HTTP 500 is returned, and confirms readiness still returns 200.

Before / After

Scenario	Before	After
Bad args to function	Worker exits, request hangs, server appears down	HTTP 500 returned, server stays up
Worker exits mid-call	Pending requests hang indefinitely	All pending requests get HTTP 500 immediately
Deploy promise after exit	deployReject called on settled promise (silent error)	Guarded by isDeploySettled flag