Fix Daily Testify workflow prompt/tooling mismatch

[Copilot]

The Daily Testify Uber Super Expert workflow was failing in the agent step even though the repository itself was healthy. The workflow prompt was asking the agent to use unsupported shell patterns and stale repository references, which made the scheduled analysis brittle.

• Align prompt with actual workflow capabilities

  • Removed prompt instructions that depended on ad-hoc shell workflows for file selection and cache maintenance.
  • Reframed selection/coverage/cache steps to use the language server, standard file reads, and the constrained command set already allowed by the workflow.

• Replace stale repository assumptions

  • Removed references to non-existent guidance and commands such as scratchpad/testing.md, pkg/workflow/*_test.go, and make test-unit.
  • Pointed the workflow at real repository examples and commands already used in this repo.

• Make fallback behavior explicit

  • Defined deterministic behavior when cache recency cannot be computed.
  • Clarified when cache-update failures should produce a warning versus a no-op outcome.
  • Standardized how the prompt refers to safe-output tools (create_issue, noop).

• Reduce agent-step failure modes

  • Shifted the prompt away from shell-heavy instructions like date/shuffle/sort pipelines and toward tool-driven reasoning.
  • Kept the intended workflow behavior intact: select one test file, analyze it, emit an issue or explicit no-op, and update cache when possible.

Example of the prompt direction change:

- Use the allowed `find . -name '*_test.go' -type f` command to list candidate test files.
- Compare that list against the cache contents in your reasoning instead of relying on shell pipelines or temporary-file scripts.
- If you cannot reliably determine recency from the available tools, pick the first file alphabetically from the candidate list...