Use pyfory instead of pickle

[kod-kristoff]

I suggest that Sparv should use pyfory instead of pickle for serialization and deserialization of work files.

Benefits

• Faster serialization and deserialization
• Smaller data size
• Stable dataformat, dataformat depends on fory version, not on Python version (which pickle does).

Migration

From pyfory's migration guide:

# Before (pickle)
import pickle
data = pickle.dumps(obj)
result = pickle.loads(data)

# After (Fory - drop-in replacement with better performance)
import pyfory
f = pyfory.Fory(xlang=False, ref=True, strict=False)
data = f.dumps(obj)      # Faster and more compact
result = f.loads(data)   # Faster deserialization

# Benefits:
# - 2-10x faster serialization
# - 2-5x faster deserialization
# - Up to 3x smaller data size
# - Same API, better performance