feat: small blob fixes/improvements

[MirandaWood]

Wee PR to fix some things I discovered in #11632:

• We don't actually support multiple blobs per block in ts. The call in data_retrieval assumes we have just one per block and doesn't push the blob buffers together. I added a fn to combine them.
• Also, the archiver.test didn't support multiple blobs and would fail when we had more than one; I edited the helper fns.
• I removed the range checks in components.nr for a (small) gate count improvement.
• In components.nr we were checking the incorrect index for public and private logs:

let index = offset + j * PUBLIC_LOG_SIZE_IN_FIELDS + k;
if index < array_len {
                    assert_eq(tx_effects_hash_input[index], log[k]);
                    assert_eq(tx_effects_hash_input[index], log[k]);
                }

Where we should be checking whether j * PUBLIC_LOG_SIZE_IN_FIELDS + k is less than array_len - no inclusion of offset.

• I noticed that we were sometimes spending 100k+ gates on array_merge. I changed it to construct the array in unconstrained, then check each item against either the left or right array. We could also use is_empty on each item, but depending on T that could be expensive, and I don't think we have a use case where we must check each item for emptiness.

[Maddiaa0]

thanks for the yarn-project fixes! ( thats all i looked at )

[github-actions]

Changes to circuit sizes

Generated at commit: 7772b601bab2fc8138f9d9dc364908f273f68ad5, compared to commit: 7124664fe90533079548d393b5df2f4c461e9446

🧾 Summary (100% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
rollup_base_private	-6 ✅	-0.00%	-590 ✅	-0.04%
rollup_root	-24,164 ✅	-78.42%	-84,229 ✅	-0.31%
rollup_block_merge	-24,164 ✅	-78.38%	-84,229 ✅	-5.00%
rollup_base_public	-120,592 ✅	-58.35%	-370,468 ✅	-24.06%

---

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
rollup_base_private	137,830 (-6)	-0.00%	1,637,359 (-590)	-0.04%
rollup_root	6,650 (-24,164)	-78.42%	26,855,310 (-84,229)	-0.31%
rollup_block_merge	6,666 (-24,164)	-78.38%	1,599,161 (-84,229)	-5.00%
rollup_base_public	86,092 (-120,592)	-58.35%	1,169,561 (-370,468)	-24.06%

[TomAFrench]

This function seems a little weird in general to me as it assumes that the input arrays cannot be more than half filled, i.e. that array_length(array1) + array_length(array2) <= N.

Previously if we had array_length(array1) + array_length(array2) > N then the circuit would be unsatisfiable but this new implementation will drop any trailing elements from array2. We should add an explicit assertion on the two lengths now.