[ios] Don't use negative content insets

[friedbunny]

When a map view was smaller than the entire viewport, negative content insets would be applied. This had the effect of making the visual "center" of every map the center of the viewport, ignoring the map view’s true center. Negative content insets would have only been valid if the map view extended outside of its frame, which cannot happen.

Applying negative content insets would erroneously offset showAnnotations:, setCenterCoordinate:, “follow” user tracking mode, and so on.

Fixes #4440.

9e7aad7d00ef09cf74f6d08f4d1e4189d613ac5a also slightly tweaks annotation padding behavior so that smaller views use proportional padding.

/cc @1ec5

[1ec5]

This should be expressed with the MIN macro for brevity. And since it's only used in one place, it should be inclined.

[1ec5]

Of course I mean the MAX macro. Definitely keep the comment. 😄

[friedbunny]

I will incline the MAXes. 😉

[friedbunny]

Force pushed these changes at 0b3c67d.

[friedbunny]

The iOS build on Bitrise is failing with xcode test exit code: 65 (and nothing else). 😞

[friedbunny]

Seeing tests fail intermittently on my machine, too:

Failing tests:
    -[MapViewTests testContentInsetsWithTinyMapView]
** TEST FAILED **

make: *** [itest] Error 65

[friedbunny]

Tests pass reliably (five in a row on Bitrise) after switching to KVO in 9e91e6c886dc15257093d032df3f6c83e947df84, which I rebased back into 183f55e.

[1ec5]

Five seconds seems rather generous. Is it necessary to make this test take 10 seconds for a frame change that should be instant?

[friedbunny]

I’ll lower this to 1.0.

This is mainly a buffer against slow CI machines. waitForExpectationsWithTimeout:handler: should never add more than a tick or two to the execution time, no matter the timeout length — if it exceeds the timeout, it’s a failure and everything stops.

I think xcpretty is swallowing our test output and the elapsed time per test on Bitrise, but locally this test completes in the blink of an eye.

[friedbunny]

→ fde8a4c#diff-64665744fc510b27e594fbb086ae26d5R438

[1ec5]

Merged as d20795b and 593f3d5.

[friedbunny]

This PR got jumped by another into release-ios-3.2.0-android-4.0.0 and I neglected to force-push again into this branch. Merged per above.