feat: code-grader plain-text fallback + workspace env preflight

[christso]

Summary

Fixes #1207 and #1208.

`code-grader` plain-text fallback (#1207, #1210)

`code-grader` now works without the JSON protocol for simple pass/fail checks. The exit code determines the score and stdout becomes the assertion text:

Exit code	Score	Verdict
0	1.0	pass
non-zero (no stderr)	0.0	fail

#!/bin/bash
# check-pages.sh
pages=$(pdfinfo report.pdf | grep Pages | awk '{print $2}')
if [ "$pages" -ge 5 ]; then
  echo "PDF has $pages pages (≥5 required)"
else
  echo "PDF has only $pages pages (<5 required)"
  exit 1
fi

Silent one-liners work too:

- type: code-grader
  command: ["bash", "-c", "[ $(wc -l < output.txt) -ge 10 ]"]

For numeric scores or multi-aspect results, the existing JSON protocol is unchanged:
```bash
echo '{"score": 0.75, "assertions": [{"text": "relevance score", "passed": true}]}'
```

Scripts writing to stderr on non-zero exit still surface as errors (existing behaviour).

Design note: initial implementation used a `shell` grader type and then a string/numeric stdout interpretation. Both were dropped after reviewing how promptfoo handles this — no framework parses plain-text strings as scores, the right boundary is exit code (binary) vs JSON (everything else). See #1210 for the design discussion.

`workspace.env` preflight checks (#1208)

Declares required system dependencies checked once before `before_all` hooks. Fails immediately with a clear diagnostic — so a 30-minute eval doesn't burn time before hitting a missing `ffmpeg`.

workspace:
  env:
    required_commands: [ffmpeg, pandoc]
    required_python_modules: [PIL, openai]

Red/Green UAT Evidence

#1207 Red: `bash -c "[ 14 -ge 5 ]"` as code-grader → score 0 (empty stdout wrongly failed)

#1207 Green:

✅ exit-code-pass   | score 1, assertion "exit 0"
⚠️ exit-code-fail   | score 0, assertion "exit 1"
✅ stdout-pass      | score 1, assertion "PDF has 14 pages (≥5 required)"
⚠️ stdout-fail      | score 0, assertion "PDF has only 3 pages (<5 required)"

#1208: eval with missing `nonexistent_command_xyz_abc` → immediate setup error before any test runs ✅

Test plan

• 8 unit tests for `code-grader` plain-text fallback
• All 2314 tests pass
• CI green ✅

Closes #1207
Closes #1208
Closes #1210

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	a5287c6
Status:	✅  Deploy successful!
Preview URL:	https://19f827ef.agentv.pages.dev
Branch Preview URL:	https://feat-1207-1208-shell-grader.agentv.pages.dev

View logs