refactor: improve price cap configuration

[ali-bahjati]

The old max cost for total transaction didn't make sense as everything is handled in the per compute unit pricing and in the grand scheme of things we care about the price of an update not a transaction (which can have between 1 to 12 updates). This will make the behaviour simpler. Also, there were cases where the tx had little updates and the per compute unit price went very high and it impacted other publishers.

This is a non-breaking change because the config reader ignores incorrect fields (I checked it).

The default cap is also changed to 10 million as there have been spikes up to 6 million in the past. If publishers have working ws + short pulls. For this reason I changed the default for pull duration to 5 seconds which can have bad impact on Pythnet but the network load is not that high.

I considered injecting slot data via websocket and polling to make sure the global data is up to date but there is no way to subscribe to latest confirmed slot and also relying on the context for the updates won't be useful (because if there is no update coming, we don't know whether it is a websocket problem or not any actual update).

After this rollout, we won't have massive sol burn incident anymore because we always halve the median of the minimum landed tx over the past slots (whereas previously we just used that value).