simplify when block_sums and spent_index are added to the db

[antiochp]

Store spent_index when calling apply_block() and block_sums when calling verify_block_sums().

This lets us simplify process_block and add_block and makes it a little clearer when/where these are added to the db (via the current batch). We do not need to pass the data around and return it from various fns.

Conceptually we have 2 (nested) places where we commit data to the db -

1. We commit the block itself to the db if we process the block successfully (fork or otherwise)
2. We only commit the following data iff this block also increases total work -
   • chain head
   • various index entries

We do this via a child batch - this child batch includes the update to chain head, indices etc. and is rolled back along with the MMR changes if the block does not increase total work.
We apply everything transactionally (db + MMRs via txhashset extension) then discard this transaction.
The outer db transaction is then committed to commit the block itself to the db (unless validation failed or other error).

This PR moves the block_sums and spent_index into the child batch. This is safe to do we rebuild both for all blocks along a fork being processed. And we only make assumptions about them being present in the db for the main chain itself.

[jaspervdm]

Still relatively new to this part of the code so it took me a bit but I've convinced myself this is correct, so 👍. Just 1 small suggestion

[antiochp]

Related - #3267 (in terms of the outer and child lmdb transactions).