Don't fail completely if we can't read a single register

[canton7]

This might be a bit tricky to implement... But:

1. If we fail to read a batch containing a register, mark them all as "maybe bad"
2. If we fail to read a single register marked "maybe bad", mark it as "bad", and always read individually
3. If we succeed in reading a register marked "maybe bad", mark it as "good" (and include in future batches)

[canton7]

I started this a while back: https://github.com/canton7/foxess_modbus/tree/feature/continue-on-register-fail. Should probably get this over the line...

[canton7]

I was thinking about that this morning, and I think the approach I was taking won't work, as it only tracks whether a register we're intending to poll is actually unavailable, and not whether a register we're spanning as part of a larger read isn't available (which does happen).

I think we need to keep a separate list ranges of unavailable registers (as a list of tuples of (start, end)). If we do a read which fails due to "Invalid Address", we break that range down into a lot of single register reads, and construct our ranges of unavailable registers from the results. That's then fed into the range construction, in a similar way to how invalid_register_ranges is handled now, except that we don't assert if a know register falls into such a range.

We should then use issue_registry.async_create_issue to create something to display to the user, instructing them to read a particular wiki page. This will have instructions on making sure the correct firmware version is selected, and to open an issue with details if it is.