to_sql misses an existing table when the name has uppercase letters

[laughingman7743]

Problem

pyathena.pandas.util.to_sql() does not find an existing table whose name was given with uppercase letters, so if_exists="fail" does not fail and data is appended twice.

to_sql checks existence with:

SELECT table_name FROM information_schema.tables
WHERE table_schema = '{schema}' AND table_name = '{name}'

It passes schema and name through unchanged. Athena stores names in lowercase and reports them in lowercase in information_schema, so name="MyTable" never matches the existing mytable.

Reproduced against live Athena on master (3c69daf):

to_sql(df, "PyAthenaToSqlCase…", conn, location, schema="default", if_exists="fail")  # creates the table
to_sql(df, "PyAthenaToSqlCase…", conn, location, schema="default", if_exists="fail")  # no error
# SELECT count(*) -> 2   (one row written twice)

• if_exists="fail" writes Parquet into the existing table's location and runs CREATE EXTERNAL TABLE IF NOT EXISTS, which is a no-op, so the rows are appended instead of raising.
• if_exists="replace" follows the same path: the existing table is not dropped and its S3 objects are not deleted, so it appends too. This follows from the code and was not reproduced separately.

The existing test_to_sql uses a lowercase generated name, so it cannot see this.

Same query, same defect class

• The check runs through conn.cursor() without disabling query result reuse. A connection with result_reuse_enable=True can get a cached "no rows" answer from before the table existed, and the existence check fails the same way. Fix metadata reflection throttling and reuse listed metadata #777 turned result reuse off for the dialect's information_schema fallback for this reason.
• schema and name are interpolated into string literals without escaping '.

Proposed fix

Lowercase and quote-escape both identifiers in the existence check, the way _columns_from_information_schema does in the SQLAlchemy dialect. Execute the check with result_reuse_enable=False. Add a regression test that runs to_sql twice with a mixed-case name: once expecting if_exists="fail" to raise, and once expecting if_exists="replace" to leave one copy of the rows.

Out of scope

• In AwsDataCatalog, information_schema filters by Lake Formation rather than failing, so a table the caller cannot see is not found either. This is the same limit Read the dialect's own queries through an API cursor and resolve federated absence from information_schema #798 documented for the dialect.
• The DDL, DROP TABLE and ALTER TABLE statements quote identifiers with backticks. Athena resolves those case-insensitively, so they are unaffected.

[laughingman7743]

Correction to the "Same query, same defect class" section: the result-reuse concern does not hold. It was measured in ENV.work_group, where result reuse works: the existing test_cursor_query_result_reuse query came back ReusedPreviousResult=True on its second run, but the same information_schema.tables query run three times was never reused. A cached "absent" therefore cannot answer the existence check, and a regression test for that case passed with and without the change. The fix is limited to the case folding. Literal quote-escaping stays because it is part of building the lowercase literals.