fix: parse rm flags correctly to avoid pathname false positives

[lunareed720]

Fixes #28

Problem

The pre_tool_use.py hook sometimes blocks safe rm commands because it regex-matches -...r... inside pathnames (e.g. soft-hold-enrollment) as if it were an rm -r flag.

Fix

• Parse the shell command with shlex.split() and interpret -r/-R/--recursive and -f/--force only when they are actual option tokens.
• Only treat truly dangerous targets as dangerous when -r is present (/, ., .., ~, wildcards).
• Keep conservative policy: any rm with both recursive+force remains blocked.

Test

Added a small unittest file under tests/ that reproduces the reported false positive and covers a few core cases.

Run locally:

python3 -m unittest discover -s tests -p 'test_*.py'