BUG: improve error message for non-string dtype with DB-API connection (GH#61385)

[adebayopeter]

Closes #61385

Problem

When using a raw DB-API connection (e.g. sqlite3 without SQLAlchemy),
to_sql(dtype=...) requires SQL type strings. Passing a SQLAlchemy type
object (e.g. DOUBLE()) raised an unhelpful error:

ValueError: A (<sqlalchemy.sql.sqltypes.DOUBLE object at ...>) not a string

This gave no indication of why a string was required or what to do
about it.

Fix

Per @rhshadrach's guidance on the issue (no automatic fallback
conversion — that logic shouldn't be owned by pandas), this PR only
improves the error message to explain the constraint and point users
toward using a SQLAlchemy engine if they need SQLAlchemy types:

ValueError: Column 'A' has dtype '...' which is not a string. When 
using a DB-API connection (e.g. sqlite3), dtype values must be SQL 
type strings (e.g. 'TEXT', 'FLOAT'). To use SQLAlchemy types, use a 
SQLAlchemy engine instead.

Tests

Added test_sqlite_dtype_not_string_raises covering the new message
on a raw sqlite3 connection.

[adebayopeter]

Thanks for the thorough review @rhshadrach — all addressed:

1. Removed the SQLAlchemy-specific suggestion from the message, since it's a non-sequitur when my_type isn't actually an SQLAlchemy type.
2. Changed "dtype" to "type" in the column-level message, per your suggested wording — agreed that "dtype" implies a pandas/numpy dtype specifically, which this isn't.
3. Removed test_sqlite_dtype_not_string_raises — you're right that it was redundant with test_sqlite_test_dtype, which already covers this exact path and now asserts against the updated message.

Pushed the latest commit. 616 sqlite tests passing, pre-commit clean.