evaluator: undocumented (and inconsistent?) bounds on bounds

[nicuveo]

What version of CUE are you using (cue version)?

the playground (v0.5.1-0.20230413144408-f681271a38ec)

What did you do?

a: >(<=3)
b: <(<=3)
c: <(<3)
d: <=(<3)

What did you expect to see?

AFAICT, this behaviour is undocumented. My expectation would be for the four lines to resolve to the following errors, respectively:

invalid operand (<=3) ('>' requires concrete value)
invalid operand (<=3) ('<' requires concrete value)
invalid operand (<3) ('<' requires concrete value)
invalid operand (<3) ('<=' requires concrete value)

This would be consistent with how most other unary operators (such as ! and -) require concrete values.

What did you see instead?

a: number
b: <3
c: <3
d: <3

What bothers me with this one is that the result feels inconsistent? For a, i interpret the bound as "must be greater than a value that is less than 3", which, yes, is any value of the type, hence number. Similarly, for b, the bound can be interpreted as "must be strictly smaller than a number that is smaller or equal to 3", therefore strictly smaller than 3. But the same logic does not hold for c for integers.

Shouldn't this behaviour be documented as part of the language reference?

[myitcv]

In the interests of getting v0.6.0 out, and because we might end up with duplicate work on v0.6.0 and v0.7.0 for this issue, pushing to v0.7.0 for now.