Add constraint support

[omus]

Adds basic constraint support for Legolas schemas so you can define constraints on the relationships between fields. Previously, this could be worked around by adding such logic into a field:

julia> using Legolas: @schema, @version

julia> @schema "old.constraint" OldConstraint

julia> @version OldConstraintV1 begin
           a
           b = begin
               @assert a == b
               b
           end
       end

julia> OldConstraintV1(; a=1, b=2)
ERROR: ArgumentError: Invalid value set for field `b` (2)
Stacktrace:
 [1] OldConstraintV1(; a::Int64, b::Int64)
   @ Main ~/.julia/dev/Legolas/src/schemas.jl:545
 [2] top-level scope
   @ REPL[4]:1

caused by: AssertionError: a == b
Stacktrace:
 [1] macro expansion
   @ ./REPL[3]:4 [inlined]
 [2] OldConstraintV1(; a::Int64, b::Int64)
   @ Main ~/.julia/dev/Legolas/src/schemas.jl:580
 [3] top-level scope
   @ REPL[4]:1

Unfortunately, the old approach would produce an extra ArgumentError which could create confusion. Additionally, when utilizing more complicated constructor logic for b these kinds of constraints can be less clear than declaring them separately like so:

julia> using Legolas: @schema, @version

julia> @schema "new.constraint" NewConstraint

julia> @version NewConstraintV1 begin
           a
           b
           @check a == b
       end

julia> NewConstraintV1(; a=1, b=2)
ERROR: Legolas.CheckConstraintError: a == b
Stacktrace:
 [1] macro expansion
   @ ./REPL[3]:4 [inlined]
 [2] NewConstraintV1(; a::Int64, b::Int64)
   @ Main ~/.julia/dev/Legolas/src/schemas.jl:619
 [3] top-level scope
   @ REPL[4]:1

Closes #69

[omus]

GitHub actions is currently having an outage

[jrevels]

xref #69

I'd like to use @check for this instead of @assert, and have it throw a Legolas-defined CheckConstraintError as laid out in the xref'd sketch

want to avoid teaching users that it's literally the same as Julia's built-in@assert, as we might add more sugar/behaviors later to this (e.g. #20)

[omus]

I'd like to use @check for this instead of @assert; want to avoid teaching users that it's literally the same as Julia's built-in@assert, as we might add more sugar/behaviors later to this (e.g. #20)

Good call. I started off with @assert just to prove this was a viable approach. I will update this PR to use @check

[kleinschmidt]

I'm generally good with this. I think a bit of explaining about the line number node manipulations would be helpful for future maintainers.

How reliable is the line-number manipulation across julia versions? I have a vague sense that it might uh not be the most reliable thing but I have nothing to back that up...

[omus]

How reliable is the line-number manipulation across julia versions? I have a vague sense that it might uh not be the most reliable thing but I have nothing to back that up...

There's a test for it and I've found it to be quite reliable so far

[omus]

I'll be waiting until #123 is merged before merging this.

[omus]

#123 has been merged. I need to update the version here and finish some downstream testing I started.