Batch Writes and Add Backpressure in Local Tx Log

[alexanderkiel]

Closes: #3886

[codecov]

Codecov Report

❌ Patch coverage is 97.95918% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 96.45%. Comparing base (2d27a50) to head (180a490).

Files with missing lines	Patch %	Lines
modules/db/src/blaze/db/tx_log/local.clj	97.24%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3894      +/-   ##
==========================================
+ Coverage   96.43%   96.45%   +0.01%     
==========================================
  Files         437      437              
  Lines       27452    27538      +86     
  Branches      631      634       +3     
==========================================
+ Hits        26474    26561      +87     
- Misses        471      472       +1     
+ Partials      507      505       -2     

Files with missing lines	Coverage Δ
modules/db/src/blaze/db/node.clj	99.27% <100.00%> (+0.05%)	⬆️
modules/db/src/blaze/db/tx_log/local.clj	97.18% <97.24%> (-1.66%)	⬇️

... and 3 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.

[alexanderkiel]

Load test results, baseline (main, 03855d9b1) vs. this branch, both with DB_RESOURCE_STORE_KV_THREADS=64:

System	VUs	Req/s before	Req/s after	Speedup	q99 before	q99 after
A5N46	32	145.7	1000.0	6.9×	693 ms	42 ms
A5N46	64	162.6	2064.3	12.7×	1781 ms	42 ms
LEA47	32	443.6	1392.5	3.1×	218 ms	31 ms
LEA47	64	452.2	1370.3	3.0×	434 ms	68 ms
LEA79	64	5617.3	5381.2	0.96×	15.6 ms	13.0 ms

The gain tracks fsync latency: on A5N46 (5.4 ms syncs) the baseline serialized one tx-log fsync per transaction, while batching amortizes it across everything buffered in flight — the median holds at ~30 ms from 4 to 64 clients and throughput still doubles with each doubling of clients at 64, i.e. unsaturated. LEA79 syncs in 8 µs, was never fsync-bound, and stays at parity (−4% at 64 VUs, within noise).