v1.18: Combine builtin and BPF compute cost in cost model (backport of #29)

[mergify]

Problem

feature gate solana-labs#30620 has been activated everywhere, SVM now consumes compute units uniformly for both builtins and BPF instructions. Cost Model should no longer estimate their cost separately.

Original PR: solana-labs#32080

Summary of Changes

• combine builtin and bpf cost into programs_execution_cost, update metrics reporting as well.
• update tests

Fixes #

---

This is an automatic backport of pull request #29 done by [Mergify](https://mergify.com).

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 95.65217% with 3 lines in your changes are missing coverage. Please review.

Project coverage is 81.6%. Comparing base (04356e7) to head (1786258).

Additional details and impacted files

@@            Coverage Diff            @@
##            v1.18     #271     +/-   ##
=========================================
+ Coverage    80.6%    81.6%   +1.0%     
=========================================
  Files         827      827             
  Lines      224496   224486     -10     
=========================================
+ Hits       181015   183264   +2249     
+ Misses      43481    41222   -2259     

[apfitzge]

change lgtm, same as the change on master branch. diff for backport is a bit large but I'm okay with it, as much of it is simply renames or simple combination of 2 fields into a single one.

Please still get approval from @t-nelson, as I was the reviewer on original PR.

[willhickey]

I'm hesitant to backport to v1.18 right now, since we're planning to recommend it for mainnet-beta in the next couple of weeks.

Do you know how much block space gets wasted, on average, by the current implementation?

[tao-stones]

The double counted built-ins CUs waste about 10-14% block space (last time I checked). Mango dashboard visualizes it nicely: the unfilled space in BLOCK FILL RATIO column can largely be contributed to this. Probably also worth pointing out: the more votes (turquoise portion in BLOCK CONTENT column), the higher un-filled ratio.

I'd like to see this being deployed to mainnet-beta, but I'd also agree it is not as critical. Happy to close this it if deemed so.

[tao-stones]

Not going to bp to 1.18