Register the SIP queue thread before pjsua_create()

[PetrShtuka]

Stacked on #35 — review that one first.

Problem

Starting the engine aborts the process when something else has already initialized PJLIB:

Assertion failed: (!"Calling pjlib from unknown/external thread. You must register external threads with pj_thread_register() before calling any pjlib functions."), function pj_thread_this, file os_core_unix.c, line 977.

pjsua_create() opens with pj_init(), which registers its caller only when it is the call that brings PJLIB up. If the host — or a second PJSIP-based stack in the same process — already initialized PJLIB, pj_init() only bumps its reference count and the thread running the start stays unknown to PJLIB. The first per-thread call inside pjsua_create() then aborts. A push-driven start runs on a transient sipQueue worker (-wakeRegistration → -startWithError:), so this is reachable from an incoming call.

ensurePJThreadRegistered() could not cover it: it returns early while the library's own gPJInitialized flag is still false, and it has to, because pj_thread_is_registered() is not safe before PJLIB is up.

Changes

• -startEngineLocked initializes PJLIB itself, registers the calling thread, and only then calls pjsua_create(). Both failure paths release that reference.
• The teardown releases it after pjsua_destroy() has released the one pjsua_create() took (confirmed against the vendored binary: pjsua_destroy2 ends with pj_log_pop_indent and pj_shutdown), so PJLIB stays up for whoever else initialized it.
• Two regression tests: a push-driven start with credentials, and a start with PJLIB already initialized by a second consumer. The second aborts the test process without this change.

Testing

./Scripts/run-package-tests.sh — 84 registry tests and 65 package tests, 0 failures. Reverting only the library hunks makes testEngineStartsWhenSomethingElseAlreadyInitializedPJLIB abort, which is how the fix was verified.

🤖 Generated with Claude Code