✨ Buffer Pool should save in PARQUET to reduce read overhead

[joocer]

Regardless of the format that the file is in, when serializing for the buffer pool it should be saved as a parquet file (unless we can make another, faster format), this allows us on the second touch of a file to reduce handling effort by having a faster format to deserialize and a format we can push quite a lot of selection and projection into the read of.

The readers should have a check if the first three bytes are PAR and use the Parquet reader regardless.