fix model loading on architectures where size_t is 32bits

[XapaJIaMnu]

Description

When porting marian to architectures of different width (x86_32/arm32/wasm) the binary file format would fail to load. The reason for that is that size_t on those architectures translates to uint32_t, whereas most commonly the binary models are built on x86_64. This pull request makes sure that when binary model is loaded or stored, all headers are of type uint64_t . This pull request would not change the existing functionality but would make porting to architectures of different width easier, as the programmer would not be facing cryptic errors when loading binary models.

• I have tested the code manually
• I have run regression tests
• I have read and followed CONTRIBUTING.md
• I have updated CHANGELOG.md

[emjotde]

@snukky do we have any regression tests that use the *.bin models?
Might be time to add a small test that can do something with memory mapping, at least in a unit test.

[emjotde]

Let's merge the unit test one first PR: #826

[emjotde]

Also, modify the changelog please.

[snukky]

do we have any regression tests that use the *.bin models?

@emjotde We have these that use .bin models, plus "wngt19" tests include versions for avx2 and avx512.

tests/decoder/intgemm/test_intgemm_16bit_avx2.sh
tests/decoder/intgemm/test_intgemm_16bit.sh
tests/decoder/intgemm/test_intgemm_16bit_sse2.sh
tests/decoder/intgemm/test_intgemm_8bit_avx2.sh
tests/decoder/intgemm/test_intgemm_8bit.sh
tests/decoder/intgemm/test_intgemm_8bit_ssse3.sh
tests/models/wngt19/test_model_base_fbgemm_packed16.sh
tests/models/wngt19/test_model_base_fbgemm_packed8.sh
tests/models/wnmt18/test_student_small_aan_intgemm16.sh
tests/models/wnmt18/test_student_small_aan_intgemm8.sh

[XapaJIaMnu]

I updated the changelog. As roman said, the regression tests include plenty of binarizations and model loadings and since those past, I think it's fine.

BTW @emjotde did the big MPI fixes land?

[emjotde]

RE: MPI, yes. that was the last sync.

[emjotde]

I have to take a look what is going on with binary file loading in general on windows, then this PR should be ready to go in.
Something seems weird in #826