fix: use argv-safe grep execution

[Gujiassh]

Summary

• replace shell-string grep execution in agent-claudecode.py with an argv-style subprocess call
• distinguish real grep failures from the normal returncode == 1 no-match case
• add focused regression tests for both no-match and actual grep error behavior

Validation

• python3 -m unittest tests.test_agent.AgentClaudeCodeRegressionTests.test_grep_returns_no_matches_on_exit_code_one tests.test_agent.AgentClaudeCodeRegressionTests.test_grep_surfaces_real_grep_errors

Notes

• this PR stays independent from the open nanoAgent lines #4 and #5; it only touches the ClaudeCode grep helper path and its new tests

[Gujiassh]

Follow-up note: I revalidated the PR-relevant agent-claudecode.py regression scope locally with python -m unittest tests.test_agent.AgentClaudeCodeRegressionTests (4 passed) plus python -m py_compile agent-claudecode.py tests/test_agent.py. The remaining full-file test failures in tests/test_agent.py still come from unrelated agent.py hardening gaps outside this PR diff.

[Gujiassh]

Quick sync from my side: this branch is still clean locally and unchanged relative to its fork tracking branch (0/0).

Current state I rechecked today:

• PR is still OPEN + CLEAN
• no local follow-up changes are pending from my side
• there are no active review threads to address

If this looks good on your end, it seems ready for review/merge.

[Gujiassh]

Closing this stale radar lane to keep the authored-PR queue current. I refreshed/revalidated the branch recently, but there has still been no maintainer response on this thread, so I am closing it for now instead of letting it sit indefinitely. If maintainers want this change later, I can reopen or resend it on a fresh branch.