sdk/metrics: copy attributes dict to prevent post-recording mutation

[tejasae-afk]

When a caller retains a reference to the attributes dict passed to counter.add() (or any instrument's add/record call) and mutates it after the call, the mutation silently corrupts the attributes stored on the aggregation and therefore on exported data points.

The root cause is in _ViewInstrumentMatch.consume_measurement: when no attribute key filter is active, the code assigned the measurement's attributes dict directly:

# before
attributes = measurement.attributes

That dict is then passed to _create_aggregation, which stores it as self._attributes in the _Aggregation base class. Any subsequent mutation of the caller's dict changes what gets exported.

The fix is a one-line copy:

# after
attributes = dict(measurement.attributes)

A shallow copy is enough for all attribute value types (str, bool, int, float, and their sequence variants) because those values are themselves immutable once stored.

The filtered path (when _view._attribute_keys is not None) already builds a fresh dict via comprehension, so it is unaffected.

A regression test is included that records a measurement, mutates the original dict, and verifies the exported data point still carries the original attributes.

Closes #4610.

you know the codebase far better than I do, happy to adjust anything here.

[linux-foundation-easycla]

• ❌ - login: @tejasae-afk / name: tejasae-afk . The commit (ead9de3) is not authorized under a signed CLA. Please click here to be authorized. For further assistance with EasyCLA, please submit a support request ticket.