fix(events): errorMonitor in the linked EventEmitter twin + handle-band guards (#4633)

[proggeramlug]

Fixes #4633.

Status of the issue's 12 failures

Most had been fixed by intervening work — on current main the events module was already at 67/69 (newListener/removeListener, rawListeners once-wrappers, listenerCount(event, listener), setMaxListeners validation, module-helper target validation: all passing). Two real failures remained; this PR fixes both → 69/69 (100%).

Fix 1 — errors/error-monitor

Three EventEmitter implementations exist (perry-stdlib events.rs, perry-runtime stream-emitter, perry-ext-events). The stdlib twin implements events.errorMonitor; instrumentation proved the twin that actually links into compiled binaries (even under PERRY_NO_AUTO_OPTIMIZE=1) is perry-ext-events — which had no monitor support. Ported dispatch_error_monitor into ext-events' emit/emit0: monitor listeners observe every 'error' emit before the regular listeners, once-wrappers prune, and the monitor never counts as handling (unhandled 'error' still throws).

Fix 2 — on/async-iterator-abort (SIGSEGV)

events.on(emitter, name, { signal }) validates its target by probing predicates with the raw EventEmitter handle (an id in the 0x38000..0x40000 band). is_event_target (event_target.rs) and object_ptr_from_value (node_stream_readwrite.rs) used 0x10000 floors, so the handle flowed through to a GcHeader read at handle-8 → unmapped memory. Raised both floors to the runtime-wide < 0x100000 handle convention (lldb-verified at both faults: EXC_BAD_ACCESS addr=0x37ff8).

Validation

• ./run_parity_tests.sh --suite node-suite --module events: 67/69 → 69/69, no compile fails.
• Blast-radius check on the floor change: full --module stream sweep — all 10 failures it reports reproduce identically on a pristine-main binary (4 known subclass/extends-* + 6 readable/pipe cases), i.e. 0 regressions.
• cargo test -p perry-ext-events (8/8), runtime event tests (15/15).
• Residual: the process-exit hang after a signal abort is pre-existing across the events abort family (the long-passing once/abort-cleans-pending-error hangs identically) — filed as events.once/events.on with AbortSignal: process never exits after completion (event loop parks on a leaked keepalive) #5056.

Code-only PR — version bump + changelog left for merge time.