docs(readme): modernize quick start eval and fix agentv compare example

[christso]

Summary

• Replace the README opening install/YAML block and separate "What it does" section with a shorter outcome-focused intro for AgentV's broader target scope.
• Rename "Why AgentV?" to "Why?" and tighten Core Concepts around eval suites, path-derived category, workspace context, target/model, policy, experiment as the named run condition, and concrete runs.
• Promote the current quick start authoring shape to top-level name, top-level target, top-level model, and policy; experiment remains the result/run concept, not a top-level eval YAML or defineEval() field.
• Keep public policy examples to runs, timeout_seconds/timeoutSeconds, threshold, and budget_usd/budgetUsd, matching PR [codex] Support eval policy config #1576.
• Keep YAML and TypeScript examples on assertions, including string shorthand plus contains, code-grader, llm-grader, and threshold usage.
• Clarify the Results layout around .agentv/results/<experiment>/<timestamp>/, resolved target identity subfolders such as copilot-sdk--claude-sonnet-4.6, index.jsonl, summary.json, and per-test artifacts/sidecars.
• Rename the README tree's generated source snapshot from task/ to test/ for reader-facing terminology; tracked follow-up av-2ac7 covers the compatibility-safe code/artifact rename.
• Add a compact Core Concepts Mermaid diagram covering eval suites/imports/tests, path-derived categories, workspace fixtures/graders, experiment conditions, target/model, policy, concrete runs, run artifacts, and Dashboard/compare/trend readers.
• Expand the TypeScript SDK section with both paths: evaluate() for application-owned runs and defineEval() for TypeScript eval files that AgentV runs through the normal runner.
• Update the eval-file validation schema and generated eval-schema.json so explicit type: rubrics accepts string shorthand criteria, matching runtime parsing and public docs.

Verification

• Ran bun run generate:schema from packages/core.
• Ran bun run lint.
• Ran bun test packages/core/test/evaluation/validation/eval-file-schema.test.ts packages/core/test/evaluation/validation/eval-schema-sync.test.ts.
• Ran bun test packages/core/test/evaluation/loaders/grader-parser.test.ts --test-name-pattern "string shorthand|criteria".
• Validated the earlier README quick start YAML snippet against EvalFileSchema before the later target/policy README guidance update.
• Ran git diff --check after the target/model/policy/workspace-isolation/runs README updates.
• Ran git diff --check after the name + target/model/policy README correction.
• Earlier in this PR: ran a one-case mock-provider eval and verified current generated artifacts include result.json, grading.json, metrics.json, timing.json, transcripts, and outputs.

Notes

• This README PR should be read with implementation PR [codex] Support eval policy config #1576. PR [codex] Support eval policy config #1576 currently rejects top-level experiment authoring and accepts name + target + model + policy.
• experiment remains the result namespace/run condition concept used by .agentv/results/<experiment>/<timestamp>/, compare, Dashboard, and application-owned evaluate() calls.
• The code-level rename from task/ / task_dir to test/ / test_dir is intentionally tracked separately in Bead av-2ac7 because it touches the public artifact contract and readers.

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	d0114d3
Status:	✅  Deploy successful!
Preview URL:	https://4b328def.agentv.pages.dev
Branch Preview URL:	https://christso-readme-quickstart-c.agentv.pages.dev

View logs