Precompute storage offsets

[edfelten]

This adds new affordances for accessing mapped storage in ArbOS, so that storage offsets can be computed at compile/initialization time, rather than being recomputed each time a data structure is opened. This is useful because storage mapping relies on Keccak256 hashing, which has nontrivial cost to compute.

The basic strategy is to precompute a storage map data structure, which records the hashed offsets of data objects within a storage-backed structure. Then when we open the structure we can use the precomputed storage map, rather than recomputing the hashes.

The L1 pricer and L2 pricer structures are rewritten to use the new affordances, because they are the largest and most complex storage-backed structures in ArbOS. Other data structures are not rewritten. I'm interested in getting reviewer opinions on whether to rewrite more structures in this PR.

This does not require a new format version because the layout of data in storage doesn't change. We only change when we compute the layout.

[codecov]

Codecov Report

Merging #1414 (b64bd7f) into master (aa9adec) will decrease coverage by 3.43%.
The diff coverage is 90.16%.

@@            Coverage Diff             @@
##           master    #1414      +/-   ##
==========================================
- Coverage   52.25%   48.81%   -3.44%     
==========================================
  Files         260      248      -12     
  Lines       33906    29942    -3964     
  Branches      555      555              
==========================================
- Hits        17717    14617    -3100     
+ Misses      14125    13272     -853     
+ Partials     2064     2053      -11     

[joshuacolvin0]

Went with #1767 instead, caching values at runtime has a slightly lower concern about possibly using incorrect offsets.