Split interface of StreamProcessor into two parts

[SimonHeybrock]

This change is motivated by an updated understanding of how we want to use StreamProcessor in Beamlime. In particular, we may want to add chunks "eagerly" (so expensive compute can start) instead of "lazily" (when we want to get an updated result plot). This is to reduce latency and improve throughput.

To avoid performing potentially expensive post-accumulation computation with every added chunk, I split StreamProcessor.add_chunk into two methods. I also added an allow_bypass flag. This is a useful feature that was "on" by default in the previous implementation, but during discussion with @YooSunYoung we discovered that it can have unintended consequences and is risky when used without care (see also docstring). Is it thus off by default.