CallDynamic regression

[srenatus]

opa eval and opa eval -t wasm differ for

# x.rego
package x
import future.keywords.if

foo.bar.allow if f(7)
foo.baz.allow if f(8)
qux.corge.allow if f(-1)

f(_) if true

and

# y.rego
package y
import future.keywords.if

static if data.x.foo.bar.allow

dd1 if data.x[input.a].bar.allow

dd2 if data.x[input.a][input.b].allow

When evaluating data.y with input as {"a": "foo", "b": "bar"}, opa eval -d x.rego -d y.rego -i input.json -f pretty data.y yields

{
  "static": true,
  "dd1": true,
  "dd2": true
}

whereas adding -t wasm only yields

{
  "static": true
}

Probably related to #5839

[srenatus]

The problem is that it's stopping to eval with

ref data.x[__local1__]: at least one rule has 1+ rules, break

when optimizing data.x[input.a].bar.allow (and same for the dd2 ref), where it really should not do that.