Native: optimize backend startup on Linux

[jpnurmi]

crashpad vs. native startup should be comparable

Local

$ pytest -v tests/benchmark.py::test_benchmark_backend -k "crashpad or native"
...
tests/benchmark.py::test_benchmark_backend[crashpad] PASSED
Min 1.465ms, Max 1.637ms, Mean 1.536ms, StdDev 0.079ms, Median 1.506ms, CPU 0.357ms
tests/benchmark.py::test_benchmark_backend[native] PASSED
Min 5.710ms, Max 7.042ms, Mean 6.445ms, StdDev 0.617ms, Median 6.579ms, CPU 2.677ms

CI

Note: the Y-axis is logarithmic

https://getsentry.github.io/sentry-native/

[linear-code]

NATIVE-199

[jpnurmi]

Worst offenders (~9ms startup):

• zeroing 8MB of shared memory, dominated by modules (3ms, ~30%)
• loading curl and its deps (LD_DEBUG=statistics) before sentry-daemon main() (4ms, ~40%)