core/runtime: panic if cueexperiment errors

It was noticed that bad values of CUE_EXPERIMENT don't produce errors
when using the CUE Go public API. In particular, calls to
`cuecontext.New()` seem to work even with bad CUE_EXPERIMENTS. This was
found because a testscript had a typo:

```
env CUE_EXPERIMENT=evalv0=0
exec go run main.go
-- main.go --
package main
// some code that used cuecontext.New()
```

The cause is that in two places, we don't inspect the error from
`cueexperiment.Init()`. Because there is no nice error return path,
we instead panic.

Signed-off-by: Matthew Sackman <[email protected]>
Change-Id: I71a73775ee55085341c26f6c795c3db33b7b3da9
Dispatch-Trailer: {"type":"trybot","CL":1208729,"patchset":1,"ref":"refs/changes/29/1208729/1","targetBranch":"master"}

internal/core/runtime/imports.go

@@ -59,7 +59,10 @@ var SharedRuntime = sync.OnceValue(func() *Runtime {
 	// but future evaluator versions may not be compatible at that level.
 	// We should consider using one SharedRuntime per evaluator version,
 	// or getting rid of SharedRuntime altogether.
-	cueexperiment.Init()
+	err := cueexperiment.Init()
+	if err != nil {
+		panic(err)
+	}
 	if cueexperiment.Flags.EvalV3 {
 		r.version = internal.DevVersion
 	} else {

internal/core/runtime/runtime.go

@@ -114,7 +114,10 @@ func (r *Runtime) Init() {
 
 	r.loaded = map[*build.Instance]interface{}{}
 
-	cueexperiment.Init()
+	err := cueexperiment.Init()
+	if err != nil {
+		panic(err)
+	}
 	if cueexperiment.Flags.EvalV3 {
 		r.version = internal.DevVersion
 	} else {