evalv3: structural cycle regression involving a builtin and repeated schemas

[mvdan]

# With the old evaluator.
env CUE_EXPERIMENT=evalv3=0
exec cue export

# With the new evaluator.
env CUE_EXPERIMENT=evalv3=1
exec cue export

-- input.cue --
package p

import "list"

#Schema: {
	required?: [...string] 
	properties?: [string]: null | #Schema
}

out: len([...#Schema] & list.Repeat([{
	#Schema // removing this fixes the bug!
	properties: foo: required: ["bar", "baz"]
}], 3))

As of f2c38f8:

# With the old evaluator. (0.011s)
> env CUE_EXPERIMENT=evalv3=0
> exec cue export
[stdout]
{
    "out": 3
}
# With the new evaluator. (0.033s)
> env CUE_EXPERIMENT=evalv3=1
> exec cue export
[stderr]
0.properties.foo: cannot combine regular field "required" with null:
    ./input.cue:7:25
0.properties.foo: conflicting values null and {required:["bar","baz"]} (mismatched types null and struct):
    ./input.cue:7:25
    ./input.cue:12:19
0.properties.foo: structural cycle:
    ./input.cue:7:32
0.properties.foo: 3 errors in empty disjunction::
    ./input.cue:10:25
[exit status 1]

Note that removing the embedded definition resolves the regression, which is very suspect.

Also note that this example is very similar to #3633.

[mpvl]

A slightly simplified reproducer:

import "list"

#D: {
	b: int
	a: null | #D
}

out: len(#D & list.Repeat([#D & {a: b: 1}], 1)[0])

[mpvl]

For reference, here is a reproducer that does not use disjunctions or embeddings:

#D: {
	b?: int
	a?: #D
}
out: len(#D & list.Repeat([#D & { a: b: 1 }], 1)[0])