Mentat

I have made huge revision of filtering library to map the current state. I have tried to document it as well as possible, so we can better plan future improvements. However, there is more pressing work at hand, so we should probably move it to version Future for now and come back to it later.

Yep, will need revisiting. Some of the operators are not consistent, were created as ad-hoc solutions, same applies for return types (especially values vs lists).

Food for thought:

• type coercion (if it's not list, make it one) vs explicit return types?
• less operators doing "the right thing" or more specific ones (scalar, list, set)
• syntax freedom (how clever to be in recognizing IPs, times, strings)?
• working with specific types
  • list if IP addresses is list of lists? list of ranges? what about operations?
  • how about time operations
• what to do with operation on nonexistent keys? Need to riddle queries with explicit "exists" calls, or somehow try to do the right thing? Is 3 > nonexistent? nonexistent > 3?

• review existing operator implementations, especially that they use current idioms (one of the suspects is "exists" not calling proper jsonpath)

Let's get inspiration from Python, but we have a special case, so let's keep it simple and translatable even to database languages. That might event mean that sometimes we will have to say that some result is undefined (db would do something else than direct code, or db does not support).

The library was renamed from mentat.filtering to pynspect, because it is being used outside of Mentat project it made sense for it to have a neutral name.

The pynspect library is now distributed as separate package.