FIX: Preserve all AcrosticConverter input characters

[Langning Zhang (Boulea7)]

Description

Fixes #2383.

AcrosticConverter now preserves digits, punctuation, Unicode whitespace, and non-letter characters instead of filtering them out. Alphabetic characters still follow the converter’s uppercase contract, including characters whose uppercase form expands to multiple code points.

The whitespace encoding uses explicit sentinel lines and escapes sentinel characters so decode() can reconstruct the complete converted prompt. Empty prompts now decode to an empty string.

Tests and Documentation

• Updated the focused converter tests with mixed text, sentinel collisions, Unicode uppercase expansions, non-letter uppercase mappings, and empty input.
• 15,029 passed, 120 skipped across the unit suite, excluding one unrelated ComicJailbreakDataset test that times out fetching a remote image on the unchanged base commit as well.
• Ruff format/check and ty pass for the changed files.
• JupyText not run because converter notebooks were not changed.