Add test verification workflow and safer generation - #3

_resolve_tests_dir() uses os.path.abspath/commonpath checks, which can be bypassed via symlinks (e.g., output_dir points to a symlink under problem_dir that resolves outside). That can still cause _clear_generated_tests() to delete .in/.ans files in external locations. Consider validating with os.path.realpath() (or Path.resolve()) for both problem_dir and tests_dir and/or rejecting symlinked output directories before clearing.

_check_file_count() computes expected_indices as range(1, len(numeric_indices)+1), which under-reports gaps when filenames skip ahead (e.g., 01.in and 100.in would only report missing 2). If you want to enforce contiguous numbering, expected_indices should be based on max(numeric_indices) (and possibly also validate the set starts at 1).

workflow_guard introduces state["tests_verified"] and advertises a required problem_verify_tests step, but the state is never updated in post_tool() and pre_tool() still allows problem_pack_polygon after only tests_generated. This makes the new verification step unenforced in the guard. Update post_tool() to set tests_verified based on problem_verify_tests output (e.g., data.get("passed")) and gate problem_pack_polygon on tests_verified.

The comment says this is saved to a "canonical/standard location that other tools depend on", but canonical_path is now derived from effective_name. Other parts of the codebase (e.g., Polygon packing XML uses solutions/sol.cpp & solutions/brute.cpp) still assume the standard filenames, so building with name!=solution_type can break downstream workflow. Consider either always also writing/copying to solutions/{solution_type}.cpp for backward compatibility, or updating downstream tools (e.g., problem_pack_polygon) to use the chosen name.