FIX Honor score feedback and handle zero-vector similarity

[Roman Lutz (romanlutz)]

Description

use_score_as_feedback=False was preserved in attack configuration but ignored by Crescendo and Tree of Attacks with Pruning. This change keeps prior target-response context available to the adversarial model while suppressing score values and rationales; TAP also skips the unnecessary score lookup. Default score-feedback behavior remains unchanged.

The audit also found that cosine similarity returned nan for zero vectors, so zero-norm comparisons now return 0.0. Stale PyRIT shell examples are migrated from the deprecated print-scenario alias to scenario-results and updated to match the current history output.

This does not change Crescendo refusal, backtracking, or terminal success semantics.

Tests and Documentation

• Full unit suite: 15,228 passed, 121 skipped.
• Changed analytics, Crescendo, and TAP modules: 196 passed.
• Ruff, changed-file type checks, async-suffix validation, documentation validation, and all changed-file pre-commit hooks passed.
• Added regression coverage for disabled score feedback in Crescendo and TAP, including the absence of TAP score lookup.
• Added zero-vector cosine-similarity coverage.
• Updated doc/scanner/2_pyrit_shell.md; JupyText is not applicable to this Markdown-only documentation change.