IPNI outages should not affect Spark RSR

[bajtos]

In the past several days, there have been periods of time when IPNI was returning 404 to most requests. As a result, Spark reports a very low RSR score, which in turn negatively affects SPs' reputation.

We should find a strategy how to handle the degradation of the IPNI service in such a way that does not affect the reputation of storage providers, clients and allocators.

[bajtos]

Also:

AFAICT, ATM, we simply assume IPNI works - there is no timeout and no retry logic. If we cannot get the response or parse the response body as JSON, we return a generic code IPNI_ERROR_FETCH.

We should add timeout & retry, potentially rotating between different IPNI endpoints/providers.