Revert "[stdlib] AutoreleasingUnsafeMutablePointer: Switch subscripts to _read accessors"

[Catfish-Man]

This was an ABI break, since it didn't make it into 5.0. Using _read here is unimportant, so we're just going to revert rather than try being fancy.

This reverts commit 04586e3.

Fixes rdar://problem/51503385

[Catfish-Man]

@swift-ci please test

[lorentey]

Cc @nkcsgexi -- this is not currently detected by the ABI checker

[nkcsgexi]

@lorentey Thanks for the heads-up! Filed rdar://52063421 to keep track of this enhancement.

[lorentey]

I guess adding the original getter and setting _read to @_alwaysEmitIntoClient would be a workable alternative.

I'm rather surprised that pointee._read is called from libswiftCore; apparently @_transparent doesn't do its thing for such accessors. It would be interesting to see what code is generated before/after this change.

[jrose-apple]

If it's satisfying a protocol requirement, it might not be inlined.

[jckarter]

If it's going through a protocol requirement, then the set of accessors on the concrete implementation shouldn't impact the abstract interface. @_alwaysEmitIntoClient would still be OK to give concrete code an opportunity to benefit.