Fix for SR-6849 and SR-6962 leak in NSData and NSMutableData

[mamabusi]

Test case:

import Foundation

for _ in 0..<10 {
        let data = Data(base64Encoded: "UE9TVCAvIEhUVFAvMS4wDQpDb250ZW50LUxlbmd0aDoNCg0K")!
}

Valgrind memcheck tool output before the fix (as reported in SR-6849):

==7078== LEAK SUMMARY:
==7078==    definitely lost: 360 bytes in 10 blocks

Valgrind memcheck tool output after the fix:

==20265== LEAK SUMMARY:
==20265==    definitely lost: 0 bytes in 0 blocks
==20265==    indirectly lost: 0 bytes in 0 blocks
==20265==      possibly lost: 0 bytes in 0 blocks

Summary:
This is a regression from the df3ec55 commit for the High Sierra changes.

Previously, PR: #380 had a fix for NSData leak where __CFDataGetInfoBit for __kCFDontDeallocate (then equivalent of __CFDataDontDeallocate before PR for High Sierra changes) was set on an ‘if’ condition in _CFDataInit method and it was in turn checked in the __CFDataDeallocate method of CFData during deallocation.

However, with the HighSierra PR 1338 changes now, the value for __CFDataDontDeallocate is hardcoded to 'true'
__CFDataSetDontDeallocate(memory, true);
With this, data is never deallocated and hence the leak.

[mamabusi]

__kCFMutableVarietyMask is not used anymore.

[mamabusi]

__kCFAllocatesCollectable is not used anymore.

[mamabusi]

Changed for bitmasks to integers on CoreFoundation's CFData.c

[spevans]

@swift-ci please test

[mamabusi]

@djones6 tested this fix for SR-6962 and observes the leak is not happening.

[parkera]

Thank you @mamabusi. I have a feeling this is one of those changes that looks simple but took forever to track down.

[parkera]

Approved for 4.1 branch as well.

[spevans]

@swift-ci please test and merge