fix(native): Skip scope flush during crash handling

[jpnurmi]

Came up in:

• WIP: fix(tracing): finish active trace on crash #1667 (comment)

native_backend_flush_scope writes the current scope to state->event_path so the daemon has a fresh snapshot on crash. In the crash handler, however, any scope-mutating code path (e.g. sentry__scope_apply_to_event via SENTRY_WITH_SCOPE_MUT's flush-on-unlock) invokes flush_scope_func while native_backend_except is already writing the real crash event to the same file. The intermediate writes are wasted (the final write overwrites them) and add unnecessary file I/O in crash context.

Mirror the crashpad backend's existing guard: add a crashed atomic flag on the state, set it at the entry of except_func, and check it in flush_func so the flush becomes a no-op for the duration of crash handling:

sentry-native/src/backends/sentry_backend_crashpad.cpp

Lines 265 to 270 in 82dbf13

	//
	if (!data->event_path || data->crashed.load(std::memory_order_relaxed)
	|| !data->scope_flush.compare_exchange_strong(
	expected, true, std::memory_order_acquire)) {
	return;
	}