docs: add dogfood report for v3.8.0 (#756)

* docs: add dogfood report for v3.8.0

10 bugs found across native Rust build orchestrator: missing import
edges, wrong build_meta version, embed path doubling, dataflow null
crash, engine flag ignored, and missing analysis passes. 6 issues
filed (#750-#755). WASM fallback path remains fully functional.
Rating: 4/10 due to native engine gaps in default auto mode.

* fix: address review feedback on dogfood report accuracy (#756)

- Add footnote clarifying no-op regression was measured under WASM due
  to BUG 2 making native no-op impossible (actual native regression is
  ~+29,000%, not +292%)
- Fix edge gap denominator to consistently use WASM as reference
  (-12.7% of WASM) matching the report body
- Add note explaining build time discrepancy between Section 2
  (single-run approximate) and Section 8 (formal 5-run average)
- Update PR description to use consistent 12.7% figure

* fix: clarify phase breakdown vs BUG 6 and benchmark table discrepancy (#756)

Add notes explaining that native complexity/CFG/dataflow phases execute
but fail to commit results (reconciling the phase timings with BUG 6),
and that the two full-build figures come from different benchmark scripts
with different pipeline overhead.

* fix: add missing footnotes for WASM-path benchmark figures (#756)

- Add ² footnote to Build Benchmark no-op row (55ms) clarifying it was
  measured under CODEGRAPH_FORCE_JS_PIPELINE=1 due to BUG 2
- Add ¹ marker to 1-file rebuild regression row since the footnote
  covers both no-op and 1-file rebuild figures
- Clarify Section 10.6 that regression percentages are WASM-path
  measurements and note the +29,000% native no-op regression

Author: carlos-alm