[Analytics] Deprecate old compression cache and move to using cache of CompressionMetadata

The compression cache currently caches a single byte array for the CompressionInfo.db file, this is a problem for large files as it involves allocating and garbage collecting large memory segments, but also means that every consumer of the bytes will instantiate a CompressionMetadata object and allocate an individual BigLongArray to store the chunk offsets. This is unnecessary as the CompressionMetadata is immutable and can be re-used.