I looked at the docs of bit_length and it does not state if it is before or after compression.
However since spark decompresses data on read, it is very likely it is the size before compression.
The table size is read from metadata and is compressed.
To be 100% sure, you can try with a file without compression.