Set the minimum buffer size in Group::write() to be equal to the page size

[tgoyne]

For encrypted files we always write data in page sized increments, so using a buffer smaller than the page size results in writing the same page multiple times. On arm64 Apple platforms, this was a problem for files between 16 KB and 4 MB.

[coveralls-official]

Pull Request Test Coverage Report for Build thomas.goyne_238

Details

• 1 of 1 (100.0%) changed or added relevant line in 1 file are covered.
• 76 unchanged lines in 13 files lost coverage.
• Overall coverage increased (+0.002%) to 91.808%

---

Files with Coverage Reduction	New Missed Lines	%
src/realm/sync/network/network.cpp	1	89.86%
src/realm/array_blobs_big.cpp	2	98.72%
src/realm/array_string.cpp	2	85.25%
src/realm/collection_parent.cpp	2	92.48%
src/realm/object-store/shared_realm.cpp	2	93.05%
src/realm/util/file.cpp	3	81.02%
src/realm/util/serializer.cpp	3	90.03%
src/realm/sync/noinst/client_impl_base.cpp	4	85.64%
src/realm/sync/transform.cpp	4	64.61%
src/realm/util/assert.hpp	4	87.1%

Totals
Change from base Build 2137:	0.002%
Covered Lines:	242852
Relevant Lines:	264523

---

💛 - Coveralls

[ironage]

I don't think we have tried to optimized Realm.write() for a long time and we are a bit blind here, but using the page size makes sense. Was it something that you observed yourself during manual profiling?
The change itself seems fine to me.

[tgoyne]

I was poking at the encryption code a bit to see how easy it'd be to remove the file id stuff. File::write() does some silly things so I checked if it was actually being used anywhere that it'd matter, and in the process noticed Group::write() was indirectly calling it with non-page-aligned sizes.