probes: add insecure code completion probe and detector (#444)

[manunicholasjacob]

Summary

Adds an insecure-code-completion probe + detector (new codecompletion module),
addressing #444, following the scenario approach of "Asleep at the Keyboard?"
(Pearce et al., 2021, arXiv:2108.09293).

What it does

• probes.codecompletion.InsecureCodeCompletion — five Python completion
  scenarios, each a realistic task that stops before a security-sensitive
  choice, covering CWE-502 (unsafe deserialization), CWE-327 (weak hashing),
  CWE-798 (hard-coded credentials), CWE-78 (OS command injection), and
  CWE-732 (over-permissive file permissions).
• detectors.codecompletion.InsecureCode — flags completions that use a named
  dangerous idiom for those CWEs.

Scope (honest notes)

• The paper evaluates with CodeQL + manual review; this port uses a lightweight,
  high-precision regex heuristic instead (documented in the detector) — it flags
  clearly-insecure idioms, not full dataflow.
• Python-only, one scenario per CWE (the paper also covers C/Verilog and uses
  three per CWE) — a focused first cut.
• Adds the five covered CWE identifiers to garak/data/tags.misp.tsv for
  accurate tagging.

Testing

• black clean, pylint 10/10
• probe + detector tests pass; all repo tag-format tests pass with the taxonomy
  additions; test.Blank end-to-end passes
• detector precision spot-checked: insecure idioms flag; safe equivalents
  (parameterized SQL, sha256, SafeLoader, restrictive chmod, input()) pass

Closes #444.