feat: add input validation for early error detection

[haoyu-haoyu]

Summary

• Add rfdiffusion/validation.py with validators for PDB files, contig strings, checkpoint paths, hotspot residues, and diffuser config
• Integrate validation into Sampler.initialize() and sample_init(), before GPU allocation and model loading

Currently, invalid inputs (missing PDB, malformed contigs, bad hotspot format) produce cryptic errors deep in tensor operations after the model is already loaded on GPU. This PR catches these errors early with clear, actionable messages.

Validators

Function	What it checks
validate_pdb_path()	File exists, contains ATOM records with valid coordinates
validate_contig_string()	Syntax: ranges 10-20, chain specs A5-50, no negatives, start ≤ end
validate_checkpoint_path()	Model checkpoint file exists
validate_hotspot_res()	Format A50 (chain letter + integer)
validate_diffuser_config()	T ≥ 1, partial_T ≤ T

Example error messages

ValidationError: Input PDB file not found: /path/to/missing.pdb
ValidationError: Invalid contig range: '20-10' (start > end)
ValidationError: Model checkpoint not found: models/Base_ckpt.pt. Please download models following the README instructions.

Test plan

• Missing PDB → clear ValidationError before model loading
• Empty PDB (no ATOM records) → clear error
• Malformed contig strings → clear error
• All valid existing inputs still work unchanged