Quote the path in DataFrameReader.load when no format is given

[Raaif-Yousuf]

DataFrameReader.load(path) fails to parse for essentially any path when no format= is given.

from duckdb.experimental.spark.sql import SparkSession

spark = SparkSession.builder.getOrCreate()
spark.read.load("data/basic.csv")

duckdb.duckdb.ParserException: Parser Error: syntax error at or near "/"
LINE 1: select * from data/basic.csv
                          ^

load() builds its query as f"select * from {path}" (duckdb/experimental/spark/sql/readwriter.py:119), splicing the raw path into the SQL text. Anything the parser treats as syntax ends the identifier, so a /, a space or a drive colon all break it, and on Windows the backslash does too. Only a bare filename in the working directory survives, because the replacement scan then picks it up. Every other reader and writer method in that file hands the path to DuckDB as an argument (read_csv(path), read_json(path), read_parquet(path)), which is why they are fine.

This quotes the path as a SQL string literal and doubles any embedded single quote.

One behaviour change worth calling out: spark.read.load("some_table") currently resolves as a table reference, and after this it is read as a path. That matches PySpark, where load() takes a path and spark.read.table() is the way to read a table, but it is a change.

Adds tests/fast/spark/test_spark_read_load.py covering a path with a space and a path with a single quote in it; nothing exercised load() without a format before. The class is skipped under USE_ACTUAL_SPARK, since real Spark defaults load() to parquet rather than reading by extension.

Verified on Windows with Python 3.12: both tests fail on main with the parser error above and pass with the change, and pytest tests/fast/spark is otherwise unchanged (289 passed, 5 skipped, 1 xfailed, plus one pre-existing test_all_types_schema failure that reproduces without this patch).