Refactor create_binary_nisar_mask to be less aggressive

[sssangha]

Updated the create_binary_nisar_mask function to clarify the conditions for valid and invalid data, and modified the binary mask creation logic to consider both reference and secondary swaths.

[sssangha]

• The Issue: The create_binary_nisar_mask function in util/stitch.py was incorrectly masking out broad swaths of valid data along the azimuth. This occurred because modulo-10 arithmetic was being applied directly to the raw mask values. If NISAR packed higher-order flags into the bits above the base 3-digit code (Bit 8+), it corrupted the modulo calculation and falsely flagged pixels as invalid.
• The Fix: * Applied a bitwise AND mask (& 0xFF) prior to digit extraction to safely strip out any higher-order bits.
  • Updated the validation logic to verify that both the reference and secondary subswaths contain valid data (> 0), bringing the base rigor in line with MintPy's handling.
  • The intentional pass-through of the water mask (hundreds digit) was preserved to maintain user control over water bodies.
• Impact: Eliminates arbitrary data loss along the azimuth and ensures consistent, reliable mask generation.

IFGs before vs after this patch: