Unguarded str()/repr() calls crash the app when an attribute key/value's __str__/__repr__ raises

Describe your environment

OS: Windows 11 (also reproduces on Linux — not platform-specific)
Python version: Python 3.11
SDK version: current main branch (unreleased)
API version: current main branch (unreleased)

What happened?

Three separate code paths use "best-effort stringify" as a fallback when
an attribute key/value, log exception message, or baggage entry isn't
already a primitive type. In all three, the str()/repr() call is
completely unguarded -- if the object's str or repr itself
raises an exception, that exception propagates straight out of the
telemetry call and crashes the calling application, instead of degrading
gracefully as documented.

1. opentelemetry-api/src/opentelemetry/attributes/init.py,
   _clean_attribute_value(): the function's own docstring says an
   unstringifiable value "is replaced with None", but str(key)/
   str(value) are called with no try/except. Crashes
   span.set_attribute(...).

2. opentelemetry-sdk/src/opentelemetry/sdk/_logs/_internal/init.py,
   LoggingHandler._get_attributes(): calls str(value.args[0]) unguarded
   when translating a stdlib LogRecord's exc_info into OTel log
   attributes. Crashes the extremely common
   logger.exception(...) / logger.error(msg, exc_info=True) pattern.

3. opentelemetry-api/src/opentelemetry/baggage/propagation/init.py,
   _encode_baggage_pairs(): calls str(key)/str(value) unguarded, crashing
   W3CBaggagePropagator.inject().

All three represent the same underlying problem: observability/telemetry
code should never crash the application it's monitoring, especially over
something as incidental as failing to format a debug string. A single
malformed or unusual object (a buggy custom repr, a flaky third-party
exception class, a corrupted data object) passed as a span attribute,
logged exception, or baggage value takes down the entire request/process.

I have a fix and regression tests ready for all three sites and will open
a PR referencing this issue.

Steps to Reproduce

from opentelemetry.sdk.trace import TracerProvider

provider = TracerProvider()
tracer = provider.get_tracer("t")

class Bad:
    def __repr__(self):
        raise RuntimeError("boom from __repr__")

with tracer.start_as_current_span("s") as span:
    span.set_attribute("bad", Bad())   # raises RuntimeError, uncaught

Also reproduces via the logging bridge:

import logging
from opentelemetry.sdk._logs import LoggingHandler, LoggerProvider

class BadArg:
    def __str__(self):
        raise RuntimeError("BadArg.__str__ blew up")

class BadException(Exception):
    pass

handler = LoggingHandler(logger_provider=LoggerProvider())
py_logger = logging.getLogger("repro")
py_logger.addHandler(handler)
py_logger.setLevel(logging.ERROR)

try:
    raise BadException(BadArg())
except Exception:
    py_logger.exception("something failed")   # raises RuntimeError, uncaught

And via baggage:

from opentelemetry.baggage.propagation import W3CBaggagePropagator
from opentelemetry.baggage import set_baggage

class BadValue:
    def __str__(self):
        raise RuntimeError("boom from baggage value __str__")

propagator = W3CBaggagePropagator()
ctx = set_baggage("bad_key", BadValue())
propagator.inject({}, context=ctx)   # raises RuntimeError, uncaught

Expected Result

None of the three calls above should crash. Per each function's own
documented/implied contract, an unstringifiable value should be dropped
or replaced with a placeholder, with a warning logged -- not propagate
the exception into the caller.

Actual Result

All three raise RuntimeError uncaught, crashing the calling code.

Additional context

Fix already submitted, bundling all three sites into one PR since they
share the same root cause and fix pattern.

Would you like to implement a fix?

Yes

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}

Start with _clean_attribute_value() in opentelemetry-api/src/opentelemetry/attributes/init.py, LoggingHandler._get_attributes() in opentelemetry-sdk/src/opentelemetry/sdk/_logs/_internal/init.py, and _encode_baggage_pairs() in opentelemetry-api/src/opentelemetry/baggage/propagation/init.py. Review the existing tests and add or run regression coverage for objects whose str or repr raises; done means attribute, logging, and baggage calls degrade as documented without propagating the formatting exception.