clientcore: refactor SSE retries

[lmolkova]

SSE retries should have:

• retry policy config - probably the same as generic HTTP one
• instrumentation - logging/tracing/metrics same as generic HTTP
• HTTP-client agnostic implementation
• stress tests to validate

We need to find a way to extend retry policy to support retrying long-running HTTP streams (or find other solution for SSE config/observability).

It involves some way for retry policy to know how to modify request to retry from the new position and a way to obtain retry-after from HTTP response or SSE message.