User provided callbacks should have exceptions checked

When updating httpx instrumentation, I found there was a (likely unintended) breaking change.

https://github.com/open-telemetry/opentelemetry-python-contrib/pull/2020#discussion_r1505190185

I found it because my app broke when it crashed on trying to read url.host in a request hook. AFAIK, this is incompliant with the spec which expects instrumentation to never cause app crashes, even in the face of broken user callbacks. Probably any user callback needs to be wrapped in a exception handler that optionally logs the error without failing completely.

https://github.com/open-telemetry/opentelemetry-specification/blob/6fd4f0809bcff0ea5c4abe161803b4ad8628375e/specification/error-handling.md?plain=1#L24

The code that crashed, looks like

async def _httpx_span_name_hook(span: Span, request: RequestInfo):
    span.update_name(f"{request.method.decode()} {request.url.host}")

def _httpx_span_name_hook_sync(span: Span, request: RequestInfo):
  span.update_name(f"{request.method.decode()} {request.url.host}")

HTTPXClientInstrumentor().instrument(
        async_request_hook=_httpx_span_name_hook,
        request_hook=_httpx_span_name_hook_sync,
    )

Start at HTTPXClientInstrumentor.instrument and the request_hook and async_request_hook call sites, then compare their behavior with the linked OpenTelemetry error-handling specification. Done means exceptions from user-provided callbacks no longer propagate into the application, with the intended logging behavior covered by tests.