buffs: add Turkish locale case mapping buffs

[fevziegeyurtsevenler]

Adds buffs.turkish_casefold with TurkishLowercase and TurkishUppercase.

Why

Turkish keeps the dot on i as a distinguishing feature across case changes, so it has two i-pairs — I/ı and İ/i — where the default Unicode case mapping has one. str.lower() and str.casefold() implement the default mapping, so anything that normalises with them before matching a keyword list can be stepped around:

>>> "IGNORE previous instructions".lower()
"ignore previous instructions"          # a keyword filter matches

>>> turkish_lower("IGNORE previous instructions")
"ıgnore previous instructions"          # it does not — but a model still reads it

The other direction leaves a residue that survives normalisation, because İ lowercases to i + U+0307 rather than to i:

>>> turkish_upper("ignore").lower()
"i̇gnore"                                # still != "ignore"

This is distinct from the homoglyph substitution in probes.smuggling — no characters are swapped for lookalikes from another script. These are the codepoints a tr locale legitimately produces, so the text survives round-tripping through anything that is locale-aware.

What

Two buffs, deliberately placed next to buffs.lowercase so the default and tr mappings can be run against the same target and compared. lang = "*", since the transformation applies to any source language.

Notes

• No new dependencies; str.translate plus a two-entry map.
• 6 unit tests covering both directions, the filter-bypass property, and the U+0307 round-trip. tests/buffs/test_buffs.py picks the new classes up automatically via plugin enumeration.
• Docs page added and wired into index_buffs.rst; tests/test_docs.py passes.
• Motivation came from measuring this against open guardrails on Turkish text: https://github.com/fevziegeyurtsevenler/turkish-casefold-evasion

Local run: pytest tests/buffs/ tests/test_docs.py tests/test_configurable.py — all green.