Optimize the compute speed for `SDFConverter`

[zhu0619]

Is your feature request related to a problem? Please describe.

The logic of SDFConverteris to load sdf entirely by datamol.read_sdf first, and then convert to zarr file.
It is very slow when dealing with larger SDF files.

Describe the solution you'd like

Parallelize the conversion while ensuring the pointers are correctly assigned.

Describe alternatives you've considered

• Parse and split the SDF file and paraellize the SDF loading process in batches.