working training on textures from 3D-FUTURE-model: edited dataset preparation script, added commands to readme and exported conda environment

Print full traceback when custom extension build fails.

Also allow pytorch 1.9 so that this runs against pytorch upstream
devel builds.

issues #2, #28, #35, #37, #39

- Add a justification for using uncompressed zip and uncompressed png (#22)
- Clarify class label usage with dataset_tool.py (#18)
- Disable tf32 in calc_metrics as this has not been tested.

Defaults to keeping tf32 disabled.  This is because we haven't fully
verified training results with fp32 enabled.

Fix OOM crash in data loader workers caused by docker's small default
shared memory size.