[zackees]

Fixes #4

• obs_module_load(): if either startup event fails to init, log, destroy the one made (os_event_destroy is NULL-safe), reset both pointers and return false. libobs does not call obs_module_unload() for a module whose load failed, so the unload path never waits on a NULL event.
• BrowserSource::ExecuteOnBrowser() (returns void): if finishedEvent fails to init, log and return without queueing the task, so nothing signals or waits on an uninitialized pointer.

Successful paths are unchanged.

RED evidence (reasoning, no harness): the existing tests/ harnesses need a full CEF build and none can inject an os_event_init() failure without stubbing libobs. Before this change, a failed init under obs-rust/obs-studio#2163 leaves the pointer NULL (module load) or uninitialized stack garbage (ExecuteOnBrowser), and the next os_event_signal/os_event_wait dereferences it. clang-format 22.1.3 clean.

[zackees]

CI is green (clang-format, the only check). obs-browser has no merge queue: ready for review and merge by the coordinator.

[zackees]

Coordinator review: LGTM.

• obs_module_load(): only qRegisterMetaType runs before the event init, so returning false leaks nothing; os_event_destroy is NULL-safe and both globals are reset, so a retry or the cef_started_event NULL check at line 129 stays safe.
• ExecuteOnBrowser(): returning before QueueCEFTask means no signal/wait on an uninitialized pointer; success path unchanged.
• RED: no harness in this repo can inject an os_event_init failure without a CEF build; the reasoning-based RED in the body is acceptable under the review-checkable criterion. clang-format green.

Verdict: approve; merging (repo has no merge queue).