Write via zip proxies

Based on this comment in Rec!2208 (merged), I put together an example of how using the zip/proxy machinery to efficiently fill SOA-style containers. This is Event/EventBase/tests/src/test_ZipWriting.cpp in the diff.

Because I also moved the main zip header to sit alongside the SOACollection header the overall MR diff looks awful, but the real changes are not so large. It might be better to look at individual commits.

cc: @peilian @ahennequ -- any thoughts?

The essence is that something like

auto const added = c.emplace_back<simd>();
added.setI( some_int_v );

generates a plain [contiguous] store to the end of the container, and

auto const added = c.emplace_back<simd>( mask );
added.setI( some_int_v );

generates a compress-store using the given mask.

You end up with various different kinds of proxy objects. I think this is a "zero overhead" way of filling containers in algorithms.

Type	Read semantics	Write semantics
Compress	Disabled	Compress-store to offset using mask
Contiguous	Contiguous read from offset	Contiguous write to given offset
ScalarFill	Load a single value from offset and broadcast to int_v	Disabled
ScatterGather	Gather from an int_v of offsets	Scatter to given offsets not implemented

Needs some small changes in Rec!2245 (merged) and Phys!806 (merged).