chore(deps): bump agentfield to >=0.1.73

[AbirAbbas]

Summary

Pulls in the OpenCodeProvider stderr-error-detection fix released in agentfield 0.1.73 (companion PR Agent-Field/agentfield#525).

Without this bump the Docker build keeps resolving to whatever older agentfield got cached when 0.1.67 was first pinned, so the SWE-AF model-prefix fix from #55 lands in production but the harness still silently swallows opencode Model not found / AuthenticationError / APIError cases that exit 0. Operators would once again see "Product manager failed to produce a valid PRD" with the migration prelude in stderr and no clue what's wrong.

Changes

• requirements-docker.txt — agentfield>=0.1.67 → agentfield>=0.1.73
• requirements.txt — same bump (bare-metal install path)
• pyproject.toml — was lagging at agentfield>=0.1.9, bumped to match

Verified

agentfield-0.1.73-py3-none-any.whl ships both _extract_opencode_error and _OPENCODE_STDERR_ERROR_PATTERNS:

$ pip download agentfield==0.1.73 --no-deps
$ python -c "import zipfile, glob; w = glob.glob('agentfield-*.whl')[0]; \
  src = zipfile.ZipFile(w).read('agentfield/harness/providers/opencode.py').decode(); \
  print('fix present:', '_extract_opencode_error' in src)"
fix present: True

Test plan

• pip download agentfield==0.1.73 succeeds and contains the fix.
• After merge: redeploy SWE-AF on Railway, re-trigger github buddy implement on Agent-Field/github-buddy#22, confirm any future opencode error surfaces with the actual cause in Railway logs (not buried under the SQLite migration prelude).

🤖 Generated with Claude Code