Fix miscellaneous static analyzer warnings

[1ec5]

This PR fixes the following static analyzer warnings, which were introduced in #6709, #7377, #6559, and #7300:

/path/to/mapbox-gl-native/platform/darwin/src/MGLNetworkConfiguration.m:16:9: Null is returned from a method that is expected to return a non-null value
/path/to/mapbox-gl-native/platform/darwin/src/MGLRasterSource.mm:63:23: Function call argument is an uninitialized value (within a call to 'make_unique')
/path/to/mapbox-gl-native/platform/darwin/src/MGLShape.mm:98:10: The left expression of the compound assignment is an uninitialized value. The computed value will also be garbage
/path/to/mapbox-gl-native/platform/ios/src/MGLAnnotationImage.m:60:54: Null passed to a callee that requires a non-null 1st parameter
/path/to/mapbox-gl-native/platform/ios/src/MGLAnnotationImage.m:65:10: The left expression of the compound assignment is an uninitialized value. The computed value will also be garbage
/path/to/mapbox-gl-native/platform/macos/src/MGLAnnotationImage.m:61:10: The left expression of the compound assignment is an uninitialized value. The computed value will also be garbage
/path/to/mapbox-gl-native/platform/macos/src/NSImage+MGLAdditions.mm:13:5: Returning 'self' while it is not set to the result of '[(super or self) init...]'

Working towards #7668.

/cc @frederoni @incanus

[mention-bot]

@1ec5, thanks for your PR! By analyzing this pull request, we identified @incanus and @boundsj to be potential reviewers.

[1ec5]

+[MGLAccountManager sharedManager] bails in IB because MGLAccountManager spins up lots of other machinery, including MGLMapboxEvents. MGLNetworkConfiguration is intentionally isolated from anything else, so initializing the singleton within IB is no big deal.

[1ec5]

@frederoni, any reason why comparing two UIImages for object equality isn’t good enough? (There’s a similar comparison of NSImages’ TIFF representations on the macOS side.)

[1ec5]

I see that testAnnotationImage() is failing as a result of my change. It’s true that drawing the image to a new context could make the two images unequal, but there are a number of reasons that would be the case anyways (such as different color profiles). In general, I don’t think a class’s -isEqual: implementation should be more thorough than the -isEqual: implementations of its ivars.

Edit: I misread the test. Looks like it’s doing something pretty ordinary.

[1ec5]

I had somehow forgotten that primitive-typed stack variables are uninitialized in Objective-C just as they are in C. ARC nils out only object pointers on the stack.

[kkaefer]

Is this a safe way to compute the hash?

[frederoni]

A XOR over the values could improve it but I'm ok with it because this is an abstract class and can't be used directly. All concrete subclasses include the hash of the GeoJSON dictionary.

[1ec5]

This is a fairly conventional implementation, as far as I know. The only hard requirement for a -hash implementation is that two equal objects (as in -isEqual:) must have the same hash. But feel free to improve on the various -hash implementations in this SDK in a separate PR.