This commit modernizes the codebase with the following improvements:

**Critical Fixes:**
- Replace deprecated scipy.misc.toimage with PIL.Image.fromarray
- Remove deprecated torch.autograd.Variable wrappers (PyTorch 0.4+)
- Fix scheduler.step() placement to follow modern PyTorch patterns
- Fix dangerous state dict loading that used key remapping
- Update Python version requirement from 3.6 to 3.8

**Code Quality:**
- Fix spelling errors throughout codebase:
  - normilize → normalize
  - reconsturction → reconstruction
  - schedular → scheduler
  - avarge → average
  - generataor → generator
  - thrshold → threshold
  - critirion → criterion
  - exsist → exist
  - famouse → famous
  - inilized → initialized
  - Lerned → Learned
- Remove dead/commented code
- Clean up broken function definitions

**Performance:**
- Add torch.no_grad() contexts to all validation/test functions
- Enable scheduler state saving/loading (was commented out)

**Project Structure:**
- Add requirements.txt for pip users
- Add comprehensive .gitignore file
- Update README typos and grammar

All changes maintain backward compatibility while bringing the code
up to modern PyTorch standards (1.7+).