Backport evmonly parse, app-hash and changeset perf fixes from giga-1

[masih]

Three small hot-loop optimisations from the evmonly serial-budget survey landed on giga-1 as #4234, #4236 and #4237 but were never brought to main (PLT-1302, PLT-1304, PLT-1305 under PLT-1297). All three cherry-pick cleanly; this PR carries them as-is, one commit each.

parseBlockTxs fed indices to its workers over an unbuffered channel, so with a known sender on nearly every transaction the rendezvous cost more than the decode and the parallel path ran slower than the serial one; workers now claim ranges off an atomic counter (about 4.5x faster at 8-32 workers). hashEVMOnlyResult wrote the app-hash stream to sha256 a few bytes at a time with a fresh length prefix per write; it now goes through a 32 KiB buffer with a scratch prefix, cutting roughly 7,000 allocations per block. The FlatKV changeset encoder allocated a KVPair, a key and a value per emitted pair; it now cuts all three from one slab per kind, the way EncodeMemoryStoreChangeSet already does, cutting roughly 10,500 allocations per block.

None of this changes bytes on the wire or in state: TestHashEVMOnlyResultGoldenVector and TestEncodeFlatKVChangeSetGolden* pin the hashed stream and the emitted pairs to values captured from the previous implementations, and parsed[i] was always written by index so parse order never mattered. Validated with go test -race ./giga/evmonly/ ./sei-tendermint/internal/evmonlyapp/ on this branch.

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[github-actions]

The latest Buf updates on your PR. Results from workflow Buf / buf (pull_request).

Build	Format	Lint	Breaking	Updated (UTC)
✅ passed	✅ passed	✅ passed	✅ passed	Sep 25, 2026, 4:47 PM

[codecov]

Codecov Report

❌ Patch coverage is 96.36364% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 66.53%. Comparing base (6018de1) to head (da51c02).

Files with missing lines	Patch %	Lines
giga/evmonly/flatkv_changeset.go	96.82%	2 Missing ⚠️
giga/evmonly/parser.go	91.66%	1 Missing ⚠️
sei-tendermint/internal/evmonlyapp/app.go	95.45%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4345      +/-   ##
==========================================
- Coverage   67.60%   66.53%   -1.08%     
==========================================
  Files        2192     2071     -121     
  Lines      168752   156767   -11985     
==========================================
- Hits       114087   104298    -9789     
+ Misses      54655    52459    -2196     
  Partials       10       10              

Flag	Coverage Δ
sei-chain-pr	90.22% <96.36%> (?)
sei-db	74.50% <ø> (-0.25%)	⬇️
sei-db-state-db	?

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
sei-db/common/keys/evm.go	90.19% <100.00%> (+5.98%)	⬆️
giga/evmonly/parser.go	96.42% <91.66%> (-0.13%)	⬇️
sei-tendermint/internal/evmonlyapp/app.go	85.71% <95.45%> (+0.59%)	⬆️
giga/evmonly/flatkv_changeset.go	92.59% <96.82%> (+10.09%)	⬆️

... and 122 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[cursor]

PR Summary

Medium Risk
Touches consensus app-hash hashing and FlatKV changeset bytes, but golden vectors assert identical outputs; parallel parse still writes results by index.

Overview
Backports three evmonly hot-path optimizations from giga-1 with no intended change to emitted FlatKV pairs or app hashes (golden tests lock both).

parseBlockTxs drops the unbuffered per-tx job channel for atomic range claims (parseClaim = 16 indices per worker), reducing sync overhead when many txs have known senders.

hashEVMOnlyResult streams the SHA-256 app-hash input through a 32 KiB bufio writer (evmOnlyHashWriter) instead of many small writes and per-field length-prefix allocations.

FlatKV changeset encoding builds KVPairs from pre-sized slabs (flatKVChangeSetBuilder) for pairs, keys, and values; keys use new keys.PutEVMKey into fixed buffers. Storage clears still allocate per iterated key.

Adds TestEncodeFlatKVChangeSetGolden*, TestHashEVMOnlyResultGoldenVector, and a multi-claim parse test.

^{Reviewed by Cursor Bugbot for commit da51c02. Bugbot is set up for automated code reviews on this repo. Configure here.}

[seidroid]

This backports three hot-loop optimisations: atomic range claiming in parseBlockTxs, a buffered app-hash writer, and slab-allocated FlatKV changeset pairs. Golden-vector tests pin the hash stream and the emitted pairs to their earlier values, and I found no correctness or safety issues in the diff.

Findings: 0 blocking | 0 non-blocking | 0 posted inline

Blockers

• None at the file/PR level.

Non-blocking

• None at the file/PR level.