src/node_errors: add re-entrancy guard to TriggerUncaughtException

[themuuln]

Problem

When an uncaught exception triggers TriggerUncaughtException → the JS handler (process._fatalException) → console.error → the inspector protocol (InspectorConsoleCall) throws again, V8 re-enters TriggerUncaughtException. This creates a tight infinite loop:

uv_run → uv__async_io → ThreadSafeFunction::AsyncCb →
CallbackScope::~CallbackScope → Function::Call →
TriggerUncaughtException → Function::Call →
InspectorConsoleCall → [back to TriggerUncaughtException]

Each iteration:

1. Widens the error (exception wrapped in a new exception string)
2. Saturates a CPU core
3. Compounds memory pressure (each error frame allocates)

This was observed in production with napi-rs plugins calling console.error during uncaught exception handling, but any JS handler that triggers inspector protocol activity during process._fatalException can hit it.

Fix

Add a thread_local re-entrancy guard to TriggerUncaughtException(). If the function is re-entered while already processing an uncaught exception, it:

1. Formats the exception directly via FormatCaughtException (pure V8 string ops — no JS, no inspector)
2. Prints the diagnostic to stderr with PrintToStderrAndFlush
3. Calls ABORT() to terminate the process

An RAII scope guard resets the flag when TriggerUncaughtException returns normally (e.g., env->Exit() path).

The guard lives entirely in node_errors.cc — no header changes, no Environment fields.

Fixes: #64326