Stack memory tracking improvements

[dthaler]

Example before instruction:

s[504...511].type=number, s[504...511].svalue=1311768467463790320, s[504...511].uvalue=1311768467463790320]

The instruction then sets

s[504...507].type=number, s[504...507].svalue=2596069105, s[504...507].uvalue=2596069105

Prior to this PR, after instruction, the state of s[504...507] is known but the state of s[508...511] is not.
Since it was a singleton number before the instruction, it is possible to retain knowledge of it by
splitting it off from s[504...511] rather than forgetting everything about those bytes. This results
in greater precision, and will allow a number of conformance tests, like lock_add32.data in PR #558 to pass exactly instead of going to TOP.

Fixes #566

[coveralls]

coverage: 90.381% (+0.04%) from 90.344%
when pulling f86b40a on dthaler:track-stack
into 4203e99 on vbpf:main.

[elazarg]

Does EBPF runtime guarantee little-endianness?

[dthaler]

Does EBPF runtime guarantee little-endianness?

No. But the CI/CD tests, as well as the ebpf-for-windows runtime, today only runs on little-endian machines.

[elazarg]

So endiannes needs to be a parameter, no?

[dthaler]

So endiannes needs to be a parameter, no?

Agreed. I may or may not get to this for a couple of days though, busy at a convention right now.

[dthaler]

So endiannes needs to be a parameter, no?

Agreed. I may or may not get to this for a couple of days though, busy at a convention right now.

Looking into this now. stack.yaml already (before this PR) has a bunch of test cases that assume little-endian.
So given that, it seems that it shouldn't block this PR to add more that assume little-endian.

That said, i am working on changes that allow supporting both, but that would be needed even without this PR.

[elazarg]

@dthaler this needs to be reopened

[dthaler]

@dthaler this needs to be reopened

I'm still working on changes for big-endian systems. Just slow because I have a bunch of other things going on besides this but I should finish it in the next two weeks and will open an endianness PR then.