Handle errors in `traces_sampler`

[szokeasaurusrex]

If an error occurs in a user's traces_sampler function, this can cause an unhandled exception, which may in some cases not even be reported to Sentry (see #2582).

As an example, try running the following code:

import sentry_sdk

sentry_sdk.init(
    traces_sampler=lambda _: 1 / 0,
)

with sentry_sdk.start_transaction(name="test"):
    print("hello, world!")

This program will crash with the ZeroDivisionError. In this case, the exception does get reported to Sentry via excepthook, but that will not always be the case (e.g. if the exception is not fatal).

When calling our other user-defined callbacks, such as before_send or before_send_transaction, we wrap the calls in a with capture_internal_exceptions block. You can therefore observe that the following example simply prints "hello, world!" without any errors.

import sentry_sdk

sentry_sdk.init(
    before_send_transaction=lambda event, hint: 1 / 0,
)

with sentry_sdk.start_transaction(name="test"):
    print("hello, world!")

We should likely also wrap traces_sampler calls in capture_internal_exceptions to prevent these errors from causing crashes and other undesirable behavior.