Ignoring additional fields only occurs when using non-keyword constructor

[omus]

As mentioned in the tour Legolas 0.5 will ignore additional fields passed into the constructor and does not propagate these fields to the resulting record. Something the tour fails to mention however is that only the Record(row) constructor will ignore these extra fields and the keyword constructor will throw an exception when additional fields are passed to it:

julia> using Legolas: @schema, @version

julia> @schema "example.foo" Foo

julia> @version FooV1 begin
           a::Real
           b::String
           c
           d::AbstractVector
       end

julia> fields = (a=1.0, b="hi", c=π, d=[1, 2, 3], e=true)
(a = 1.0, b = "hi", c = π, d = [1, 2, 3], e = true)

julia> FooV1(fields)
FooV1: (a = 1.0, b = "hi", c = π, d = [1, 2, 3])

julia> FooV1(; fields...)
ERROR: MethodError: no method matching FooV1(; a=1.0, b="hi", c=π, d=[1, 2, 3], e=true)
Closest candidates are:
  FooV1(; a, b, c, d) at ~/.julia/packages/Legolas/C9kX7/src/schemas.jl:500 got unsupported keyword argument "e"
  FooV1(::Any) at ~/.julia/packages/Legolas/C9kX7/src/schemas.jl:497 got unsupported keyword arguments "a", "b", "c", "d", "e"
Stacktrace:
 [1] kwerr(kw::NamedTuple{(:a, :b, :c, :d, :e), Tuple{Float64, String, Irrational{:π}, Vector{Int64}, Bool}}, args::Type)
   @ Base ./error.jl:165
 [2] top-level scope
   @ REPL[20]:1

I can see us making a distinction in the behaviour between these two constructors however we should document this difference.

Originally noticed this when reviewing some Beacon-internal generic record construction code where I thought the intermediate construction of a NamedTuple could be eliminated.