Reject duplicate label names before metric registration

[fhgffy]

Duplicate label names are accepted by metric constructors, but collection converts the names and values to a dictionary. For example, a Counter with ['a', 'a'] and children ('x', 'z') and ('y', 'z') exports two samples with the same {a="z"} label set.

Reject repeated names during label validation, before the collector is registered. This applies to Counter, Gauge, Summary, Histogram, Info and Enum. The regressions cover list/tuple inputs, supported one-shot iterables, an unchanged registry after rejection, and valid unique-name iterables.

Validation on Python 3.14:

• The final regression fails against the original validator in all 17 duplicate-name cases.
• Full suite: 419 passed, 12 skipped; 17 subtests passed.
• Flake8, the repository's pinned isort 5.10.1, git diff --check, and wheel build pass.

@csmarchbanks