Prometheus: Fix creation of invalid dataframes with exemplars

[kylebrandt]

With different labels for examplars, the prom response parser would create invalid frames (different field lengths). The prom datasource would then panic processing these frames.

This adds a test case for it and addresses the issue.

Which issue(s) does this PR fix?:

Fixes #73654

Special notes for your reviewer:

It was creating something like the following:

(*data.Frame)(0xc000b3f7c0)({
 Name: (string) "",
 Fields: ([]*data.Field) (len=4 cap=4) {
  (*data.Field)(0xc000aadaa0)({
   Name: (string) (len=4) "Time",
   Labels: (data.Labels) ,
   Config: (*data.FieldConfig)(<nil>),
   vector: (*data.timeTimeVector)(0xc0002018d8)((len=2 cap=2) {
    (time.Time) 2022-06-01 12:28:36.81 +0000 UTC,
    (time.Time) 2031-12-03 17:48:36.81 +0000 UTC
   })
  }),
  (*data.Field)(0xc000aadad0)({
   Name: (string) (len=5) "Value",
   Labels: (data.Labels) (len=9) __name__=traces_spanmetrics_duration_seconds_bucket, group=mythical, instance=f1c32fe505c3, le=0.256, service=mythical-requester, source=tempo, span_kind=SPAN_KIND_CLIENT, span_name=requester, span_status=STATUS_CODE_OK,
   Config: (*data.FieldConfig)(<nil>),
   vector: (*data.float64Vector)(0xc0002018f0)((len=2 cap=2) {
    (float64) 1.134971,
    (float64) 1.134971
   })
  }),
  (*data.Field)(0xc000aadb30)({
   Name: (string) (len=7) "traceID",
   Labels: (data.Labels) ,
   Config: (*data.FieldConfig)(<nil>),
   vector: (*data.stringVector)(0xc000201908)((len=1 cap=1) {
    (string) (len=32) "46e15ae41852149068aeb26c64643c29"
   })
  }),
  (*data.Field)(0xc000aadb60)({
   Name: (string) (len=8) "ztraceID",
   Labels: (data.Labels) ,
   Config: (*data.FieldConfig)(<nil>),
   vector: (*data.stringVector)(0xc000201950)((len=1 cap=1) {
    (string) (len=32) "46e15ae41852149068aeb26c64643c29"
   })
  })
 },

Which wouldn't marshal to JSON, so if we had the test case the existing tests would have failed.

Please check that:

• It works as expected from a user's perspective.
• If this is a pre-GA feature, it is behind a feature toggle.
• The docs are updated, and if this is a notable improvement, it's added to our What's New doc.

[toddtreece]

not really related to this PR, but one thing i just noticed while reviewing this locally is that we don't seem to cover an exemplar without labels in tests.