experiment: precise heap tags

[crusso]

Implement precise tagging of heap allocated objects.

Splits

• TAG_ARRAY into consecutive TAG_ARRAY_I, TAG_ARRAY_M, TAG_ARRAY_T, TAG_ARRAY_S (immutable array, array, tuple, shared function).
• TAG_BLOB into consecutive TAG_BLOB_B , TAB_BLOB_T and TAG_BLOB_P, TAG_BLOB_A (blob, text, principal, actor reference).
• TAG_BITS_32 into consecutive TAG_BITS32_U, TAG_BITS32_S and TAG_BITS32_F ((boxed) Nat32/Int32, Float32 (reserved for future use).
• TAG_BITS_64 into consecutive TAG_BITS64_U, TAG_BITS64_S and TAG_BITS64_F ((boxed) Nat64/Int32, Float64).

Modifies array slicing logic to preserve the base tag of the underlying array, by stealing the top 2 bits of the slice start field, relying on the fact that array indices are < 2^30. For clarity, the tag in decompressed on each iteration. For efficiency, it could actually be maintained during slicing compressed form as a sort ( see 448962f).

TODO:
[ ] print refined tags in debug output? (rts debug printing is rarely used and can be adapted later)

[github-actions]

Comparing from c899462 to 0082f1d:
In terms of gas, 4 tests regressed, 1 tests improved and the mean change is -0.7%.
In terms of size, 5 tests regressed and the mean change is +0.3%.

[crusso]

I think I've addressed all the issue. PTAL @luc-blaeser

[luc-blaeser]

Also the tagging of the actor and shared function references looks really good. Thank you very much. I will now try to merge with the EOP PRs.

[ggreif]

The patch is long, but mostly type-directed, I don't see correctness hazards. Hopefully the blob conversions will not cause perf concerns, as they are mostly small. LGTM

[crusso]

@ggreif, since Luc has already merged this branch into his EOP PRs, and a squash merge of this PR to master may cause him trouble merging later, we've decide not to merge this branch directly into master but as part of larger PR. That's also why I'm not addressing the issues you found (which are mostly cosmetic).

[crusso]

Not merging into master as already merged into Luc's EOP PRs.

[luc-blaeser]

I now merged the PR into the EOP PR stack. I think we could probably now close this PR.
@crusso: Thank you again for this great contribution!