Android 10.3.2 pre

[TimSylvester]

A special release for a particular user experiencing rare crashes in the field only. Call stacks indicate several crashes that seem like they must be caused by members from the pointer portion of key to placedVerticalIconIndex becoming unexpectedly invalid.

We insert fixed values between each member and check them on each access in hopes of narrowing down the range of potential places from which this could be caused and skip further processing of the symbol instance once any invalid value has been detected.

Also includes the benchmarking activity from #1850 to confirm that there's no major effect on performance.

The option to log events on a separate thread is extended to be a per-severity option, with the default changed to log errors synchronously but use threads for lower-severity events. This should make it more likely that relevant log events are written in the event of a crash.

10.3.1

Result(average=1137.1406652094868, low1p=206.20395674648492),
Result(average=972.6569705258048, low1p=195.68381800358537),
Result(average=1006.3422475857232, low1p=196.21148643908523),
Result(average=979.1130397712535, low1p=195.33316971831644),
Result(average=999.3013716059238, low1p=192.15926101090554),
Result(average=967.7307632612294, low1p=191.03237983679773)]})

10.3.2-pre

Result(average=1102.1602248079416, low1p=203.00198974531042),
Result(average=997.9265047694663, low1p=189.26893108387634),
Result(average=1043.7275178877694, low1p=194.42137487148483),
Result(average=1049.3060329936836, low1p=198.54736418525866),
Result(average=1036.1822841437606, low1p=192.19624842621522),
Result(average=1043.3555232138212, low1p=196.9277046632644)]})

clang-format modifies almost the entire contents of every file, so I'll apply that after people have had a chance to look at the differences.

[TimSylvester]

Intentional corruption of random SymbolInstance objects resulted in no crashes:

[louwers]

Let's not merge for now, but I can release this from this branch.