What this PR does:

When a pushed series has no __name__ label, the error Cortex returns to the
remote-write client gave no indication of which series was at fault, only
no metric name label (HTTP 500) or sample missing metric name (HTTP 400).
As the issue reporter described, the only way to find the culprit was to bisect
by stopping Prometheus instances one at a time.

This enriches both user-facing push paths so the offending series labels are
always included in the error:

• noMetricNameError now carries the series and renders it via formatLabelSet,
  exactly like every sibling validation error in pkg/util/validation/errors.go
  (tooManyLabelsError, labelValueTooLongError, the native-histogram errors,
  etc.). This is the validation path (HTTP 400) taken when metric-name
  enforcement is enabled (the default).
• The fatal error from tokenForLabels in the distributor (the
  shard_by_all_labels=false path that produced the original context-less HTTP
  500) is wrapped with the series labels via
  cortexpb.FromLabelAdaptersToMetric(ts.Labels).String().

Example, before vs after (validation path):

before:  sample missing metric name
after:   sample missing metric name metric: "{foo=\"bar\"}"

Which issue(s) this PR fixes:
Fixes #5802

Checklist

• Tests updated
• Documentation added (N/A - no user-facing config/docs change)
• CHANGELOG.md updated
• docs/configuration/v1-guarantees.md updated if this PR introduces experimental flags (N/A - no new flags)

AI assistance disclosure (per the project's GENAI_POLICY.md): this change was
prepared with the help of an AI coding assistant. I have reviewed every line,
understand the behaviour of both push paths, and validated it locally (build,
unit tests for both changed packages, golangci-lint, gofmt/goimports).