Separate verify failure templates for static and suggested

[vm-mishchenko]

Summary

Refactors verify failure feedback to use dedicated templates for static and suggested checks. Rather than wrapping failure output in a shared template, each failure type now renders its own self-contained section with a heading, intro sentence, and bulleted list of failing commands—without command output included in the prompt context.

Full command output is preserved in per-step log files and additionally appended to implement-spec.log for the failed attempt, allowing investigators to trace what triggered each retry without bloating the implementation agent's prompt.

Changes include:

• New templates: static_verify_errors.md and suggested_verify_errors.md replacing verify_errors.md
• New helper functions: _format_commands_bullets(), _render_static_verify_failures(), _render_suggested_verify_failures(), _format_static_verify_failures(), _append_failure_block()
• Updated _render_prompt() to insert verify_errors verbatim instead of wrapping with template
• Both static verify and suggested-check failure paths now render templates and append full output to logs

Test plan

• New unit tests for _format_commands_bullets() verify single and multiple command formatting
• Tests for _render_static_verify_failures() and _render_suggested_verify_failures() verify heading, intro, bullets appear and output does not
• Tests for both static verify and suggested-check timeout scenarios (rc==124) confirm output is excluded
• Integration tests confirm full output blocks are appended to implement-spec.log after failures
• Updated existing tests to expect bullets (- cmd) instead of full output in verify_errors
• Verified that test_format_suggested_failures_structure() still passes against output-included function
• Tested that _render_prompt() inserts error sections verbatim without adding wrapper heading