Update docs with latest benchmark results and blog post fixes (#78)

* Update docs with latest benchmark results and blog post fixes

- benchmark-results/index.md: All tables updated with corrected numbers,
  added GPQA thinking effort ablation section
- blog technical-deep-dive: Updated budget alternatives, algorithm comparison,
  selector summary, Opus+Opus fix, cumulative cost wording
- mkdocs.yml: Minor config updates

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Add topology assumption tradeoff for Hill Climbing vs Arm Elimination

Arm Elimination is assumption-free (uses only observed data), while
Hill Climbing requires a hand-crafted model ranking upfront. Also
split LM Proposal into its own paragraph.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Add server latency column to HotpotQA and MathQA tables

Added Avg Latency (s) column to all 6 tables (Top 15, Bottom 15,
Full 81) for both 2-tuple benchmarks using server-side latency
from cache.db results.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Fix HotpotQA Bottom 15 rank offset (was off by 1 vs Full 81)

Kimi + Haiku 4.5 is rank 66, not 67. Bottom 15 now correctly
starts at rank 67 and matches the Full 81 table.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Sort all selector comparison tables by Mean Accuracy descending

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: Sripadkarne