Improvements across process management utilities

[jeffreybulanadi]

Closes #5

Changes in this PR:

ProcessKiller

• Build the child map once in killTreesUnixMapped() and pass it to collectDescendants() instead of rebuilding it per PID
• Convert killTreesUnixFallback() from recursive to iterative using an explicit queue, preventing potential stack overflow on deep process trees
• Add missing space before array literal in return statement in buildProcMaps()

SystemUtilities

• Replace substr_count + 1 with preg_match_all('/^processor/m') for CPU count from /proc/cpuinfo, which handles edge cases around the first processor entry correctly

ExceptionHandler

• Replace assert() type narrowing with real conditional checks so behavior is consistent regardless of assertion configuration
• Reuse a single ReflectionObject instance across setExceptionLocation() and appendWorkerStackTrace() instead of creating two separate instances

ProcessSpawnHandler / ProcessManager

• Add final keyword to both classes, which are already marked @internal

BackgroundProcess

• Replace strpos() !== false checks with str_contains() in isRunning()

ProcessPool

• Use PoolShutdownException instead of RuntimeException when rejecting tasks on a shut-down pool
• Note the rejection type in the run() docblock

functions.php

• Detect objects inside arrays in emit() using array_walk_recursive, so arrays containing objects are serialized correctly instead of producing empty JSON objects

phpstan.neon

• Pin phpVersion to 8.4 so local analysis targets the same version as CI

[rcalicdan]

This looks fantastic! Thank you for following up so quickly. The implementations here are excellent and especially the use of array_walk_recursive() to solve the emit() array serialization gap (very clean and performant), and the iterative queue refactor in ProcessKiller. You addressed all the agreed-upon points perfectly while respecting the constraints we discussed. I'm approving and merging this now. Thanks again for your contribution to the Hibla ecosystem!