fix(cpp): bind native ATTACH/DETACH parameters and use SQLCipher key API

[pbbadenhorst]

Summary

Hardens four native bridge surfaces that previously concatenated user-controlled values into SQL:

• ATTACH/DETACH (SQLite, libsql, Turso bridges) — switches to bound parameters (ATTACH DATABASE ? AS ?, DETACH DATABASE ?). SQLite, libsql, and Turso all accept binding in both the path slot and the schema-name slot, so no escape logic is needed at all.
• SQLCipher key handling — switches from PRAGMA key = '<key>' to the sqlite3_key_v2 C API.
• Android CMakeLists.txt — orders cpp/sqlcipher ahead of cpp/ on the include path so the SQLCipher header (not the plain SQLite header) wins under USE_SQLCIPHER. Latent before this change because nothing referenced SQLCipher-only symbols; required now for sqlite3_key_v2 to link.
• JS bridge — rejects zero bytes in attach/detach inputs before binding.
• Adds regression tests for attach-alias injection, attach-path quoting, detach injection, and SQLCipher keys containing quotes.

Security framing

Not all of the above are security fixes; the diff is more honest if labelled:

• Real injection class fixed: quote-based payloads in the SQLCipher key ('; ATTACH DATABASE '/tmp/evil.db' AS x; --) and in ATTACH/DETACH aliases. If those inputs were ever attacker-influenced (URL, deep link, IPC payload), the pre-fix code would have executed arbitrary SQL on the connection. The C API and parameter binding kill that.
• Correctness, not security: zero-byte rejection in attach/detach. Zero-byte truncation in an attacker-controlled input only shortens the attacker's own value, so it isn't a useful exploit class. The realistic failure mode is a caller passing a path or alias that happens to contain a zero byte — silently truncated by libsql pre-fix, now rejected explicitly.
• Defense in depth, not exploit: the SQLCipher key no longer enters SQL text and is therefore not visible to trace/log callbacks.

Notes

Turso attach/detach code was updated for parity, but the example tests keep the existing Turso skip because ATTACH support is not currently exercised there.

[pbbadenhorst]

The zero byte checks might be too much - I am just paranoid doing these sort of changes.

[ospfranco]

Cool, thanks!