ref(scope): Let global scope flush itself

[jpnurmi]

A significant step toward letting users acquire direct access to the global scope.

One of the blocker problems for direct global scope access that came up in #2160 was that it would bypass flushing, which was previously controlled by the internal WITH_SCOPE wrapper macro. This PR solves the problem by assigning a flush_func to let the global scope flush itself after mutations.

[cursor]

Stale Bugbot comment from a previous run.

[codecov]

Codecov Report

❌ Patch coverage is 94.78261% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 75.55%. Comparing base (1ae39d3) to head (c77cb6e).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2167      +/-   ##
==========================================
+ Coverage   75.48%   75.55%   +0.07%     
==========================================
  Files         103      103              
  Lines       28089    28171      +82     
  Branches     5150     5159       +9     
==========================================
+ Hits        21202    21286      +84     
+ Misses       5545     5530      -15     
- Partials     1342     1355      +13     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want reviews to match your repository better? Bugbot Learning can learn team-specific rules from PR activity. A team admin can enable Learning in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit c441a85. Configure here.}

[limbonaut]

This was flagged during a scan:

Running flush_func under observers_lock can deadlock with sentry_close(). With Crashpad’s external reporter enabled on macOS or Windows, the flush calls sentry__envelope_new(), which acquires the options lock. Concurrent shutdown holds that lock and waits for observers_lock, so both threads block indefinitely. I reproduced this lock ordering with an isolated backend hook. Using sentry__envelope_new_with_dsn(options->dsn) in flush_external_crash_report() avoids the nested options-lock acquisition and lets the reproduction complete.

Suggested fix:

--- a/src/backends/sentry_backend_crashpad.cpp
+++ b/src/backends/sentry_backend_crashpad.cpp
@@ -320,7 +320,7 @@ flush_external_crash_report(crashpad_state_t *state,
     const base::FilePath &external_report_path, const sentry_options_t *options,
     const sentry_uuid_t *crash_event_id)
 {
-    sentry_envelope_t *envelope = sentry__envelope_new();
+    sentry_envelope_t *envelope = sentry__envelope_new_with_dsn(options->dsn);
     if (!envelope) {
         return;
     }

[limbonaut]

Nice improvement. Scope setters know better which changes require persistence, so the caller does not have to pick the correct macro (a potential failure point). And it helps batching. Take a look at the flagged issue above before merging.