About relaxing the validation rules, I'm not a fan of that. The point of validation is to detect potential problems. If you have false positives, you can always disable validation altogether in production environement. If you have false negatives (undetected mismatches), well... You're screwed, because you won't know about it until a runtime error occurs.
Relaxing the rules would introduce false negatives (which is not good, see above), and will make validation, indexing and projection more complex. We may consider this in 6.0 (since we'll have a more flexible foundation, which hopefully will make some parts of the code simpler), but in 5.x I'd rather not.