try_inverse has a trait bound that is too restrictive (i.e. Real)

# Summary
At the moment, the `try_inverse` function has a trait bound of Real. I can't think of any technical reasons (although there might exist some) not to just use Field.

This also might be applicable to the decomposition algorithms, although I'm not familiar with their details.

# Motivation
If the trait bound is loosened, it opens up the ability to work with matrices on different Fields. For example, I want to work with rationals, so I don't have to worry about floating point arithmetic errors.

# Drawbacks
* This might conflict with any optimizations you may have for floating point numbers, if you have any. However, this should be able to be solved with specialization.
* There might be some technical reason I'm not aware of why this won't work.