Add docs/: setup, concepts, methodology, FAQ, and codex support

[shekharprateek]

The problem

Every link in the README's documentation map was a 404 — docs/ did not exist on main. A new reader hit a dead end before reaching a first run. Separately, codex is fully supported in the harness but was invisible in the docs and omitted from the --agent list.

What this adds

16 doc pages — setup and explanation, not published results:

getting-started · hosting-paths · how-a-run-works · benchmark-your-own-repo · repository-structure · why-this-exists · vision · diagram-ascii · cost-per-task-methodology · serving-optimization-notes · omp-setup · kiro-cli-setup · codex-setup · faq/wiring-claude-code-to-models · faq/wiring-codex-to-models · faq/wiring-omp-to-models

codex-setup.md and docs/faq/ landed upstream after the first import, so they were missing here.

--agent list corrected in two places

Both omitted codex, which the harness has supported all along — 68 references in run-swe-headless.py, and run-e2e-benchmark.sh validation at line 167 already accepted all five agents.

• README.md: --agent claude|pi|omp|kiro → claude|pi|omp|kiro|codex
• run-e2e-benchmark.sh:84: the inline comment still read "claude (Claude Code) or pi (pi agent)" — it predated omp, kiro and codex entirely.

Deliberately left out (README rows removed where they existed)

Doc	Why
gpu-selection-h200-vs-l40s.md	10 links into self-hosted/vllm/benchmark-output/ — committed throughput dashboards this repo doesn't ship. It's a report on that data.
release-notes/0.1.0.md, 0.2.0.md	Both source their figures from docs/metrics/pareto-frontier-*.json and quote per-model scores and per-tier breakdowns.
harness-{pi,omp,claude-code,kiro-cli}-swe*.md (6)	Not onboarding guides. harness-pi-swe2.md opens # Results: pi harness (swe2) and is a per-model table of scores, costs and token counts. Onboarding for these harnesses is covered by omp-setup.md, kiro-cli-setup.md, codex-setup.md, docs/faq/ and benchmarks/docs/path-*.md.
results-*, agentic-coding-swe-comparison-*, best-harness-selection, model-selection-by-complexity, swe-router-evaluation*	Published model rankings.
metrics/ (12), images/ (54), slides/ (5)	The frontier JSON and rendered charts behind those rankings.

Edits to the imported pages

All to remove references to content this repo doesn't carry:

• benchmark-your-own-repo.md — its "use the frontier we already published" path pointed at an excluded comparison doc. Repointed at the reference frontier that actually ships, vend/swe-router/models.json, noting that file's own single_repository_warning.
• why-this-exists.md, cost-per-task-methodology.md, omp-setup.md, diagram-ascii.md — dropped links to excluded comparison/results docs, the slides/ directory, and two raw run-data files.
• cost-per-task-methodology.md — two sentences referred to a README leaderboard that no longer exists.
• vision.md, kiro-cli-setup.md — dropped three issue links into the aarora79 personal repo; those issue numbers don't exist here.
• repository-structure.md — the tree still said claude-code-multi-model/, omitted docs/ and vend/, and claimed a committed Hello-World example under swe-benchmark-data/ that isn't there (that path is gitignored run output).

Verified

• Every relative link in README.md resolves
• Every internal link across docs/ including docs/faq/ resolves — 28 broken ones found and fixed
• No aarora79 refs, no references to excluded results docs or docs/images
• bash -n clean on the edited script
• Pre-commit conditions clean: no trailing whitespace, no missing final newlines, nothing over the 1536 KB ceiling