Restrict identifying and non-identifying Collector attributes

We should restrict Collector-submitted identifying and non-identifying attributes. Limit the number of attributes and truncate keys, string values, and byte values before persisting them.

The extractor should skip blank keys, de-duplicate keys after truncation, and emit one summary warning instead of logging each rejected attribute.
`CollectorInstancesResource#attributesToMap` should also handle duplicate keys defensively because `Collectors.toMap` currently fails on duplicates.

Additionally, avoid logging complete untrusted OpAMP messages and health trees, since the persisted data may be bounded while log output remains large.

Start by tracing Collector-submitted attributes and the CollectorInstancesResource#attributesToMap entry point. Check how attributes are persisted and how OpAMP messages and health trees are logged. Done means bounded and de-duplicated attributes, one summary warning for rejected values, defensive duplicate-key handling, and no complete untrusted messages or health trees in logs.

Written by the indexing model from the issue text.