Previously, a fixed size buffer was allocated each time a call to
`computeSha256DigestForLargeFile` was made. This causes
undesirable performance issues. This change allocates a buffer
per apex/module-update session instead.
To fix this, that method is amended to take in a buffer passed
in by the caller instead.
To aid in the usage of this API, a new method called
`createLargeFileBuffer` is introduced to help callers allocate
a pre-determined buffer size to be passed into the SHA
computation method based on whether the device is low RAM
device or not.
Subsequent RSS benchmarking shows that this helps to
significantly lower the RSS high watermark (HWM) for
the `system_server` process into just noise.
Bug: 229137737
Test: Manual
Change-Id: I6d2d17353f999bd6cfb937cd5ae6e6925f1353ac
53abe6077d