feat(fake_data_generator): Add per-record variance to static log generation

[cijothomas]

Summary

The static log generator previously produced batches where all records were
nearly identical — same body, same attribute values, same severity, no trace
context. This made payloads unrealistically compressible by Arrow/columnar
encoders (~57:1 zstd ratio), giving misleading throughput numbers in load tests.

Changes

• Body: 50 realistic log message templates (~150 chars each) cycled across
  records. When log_body_size_bytes is configured, templates are repeated to
  fill the target size. When 0, body is omitted entirely.
• Attribute names: unified pool of 80 OTel semconv attribute names (e.g.
  thread.id, http.route, db.query.text). Overflow names use attr_N.
• Attribute values: 80-entry pool of realistic strings — URLs, UUIDs,
  hostnames, SQL queries, error messages, user agents — cycled per record.
  Special-cased types for thread.id (int), http.response.status_code
  (weighted distribution), etc.
• Severity: cycles 80% INFO / 15% WARN / 5% ERROR instead of all-INFO.
• TraceID / SpanID: random unique IDs per log record, matching real
  log-to-trace correlation behavior.
• Compression test: new test_compression_ratio_is_realistic asserts
  zstd ratio stays in 3:1–50:1 range (currently ~19:1), guarding against
  regression to the old all-identical regime.
• Config docs: updated log_body_size_bytes doc to reflect pool behavior.

Impact

Both fresh and pre_generated strategies benefit since all variance is
computed at generation time. PreGenerated retains zero runtime allocation
cost — only the pre-built batch content is more realistic.

Compression ratio improved from ~57:1 (before) to ~19:1 (after) with
log_body_size_bytes: 1024, num_log_attributes: 6.

[codecov]

Codecov Report

❌ Patch coverage is 98.05825% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 88.34%. Comparing base (d8e64e0) to head (18c6b95).
⚠️ Report is 7 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2543      +/-   ##
==========================================
- Coverage   88.34%   88.34%   -0.01%     
==========================================
  Files         613      613              
  Lines      222675   222734      +59     
==========================================
+ Hits       196731   196772      +41     
- Misses      25420    25438      +18     
  Partials      524      524              

Components	Coverage Δ
otap-dataflow	90.23% <98.05%> (-0.01%)	⬇️
query_abstraction	80.61% <ø> (ø)
query_engine	90.74% <ø> (ø)
syslog_cef_receivers	∅ <ø> (∅)
otel-arrow-go	52.45% <ø> (ø)
quiver	91.92% <ø> (ø)

🚀 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.

[cijothomas]

I was thinking of using the semantic_conventions itself, and generate a bunch of them at startup and then cycle through it - But the existing ones can only produce logs of size ~300 bytes, and we are targeting ~1KB, so we anyway have to put fillers in. This is first step in improve fake-gen, next is to have the perf tests switch to this.

[lalitb]

Nice work! Few minor nits but nothing blocking — LGTM.