Write witness caches when writing the best block
For steady-state operation, this reduces the average time between wallet disk
writes from once per block to once per hour.
On -rescan, witness caches are only written out at the end along with the best
block, increasing speed while ensuring that on-disk state is kept consistent.
Witness caches are now never recreated during a -reindex, on the assumption that
the blocks themselves are not changing (the chain is just being reconstructed),
and so the witnesses will remain valid.
Part of #1749.
Gather release notes from previous release to HEAD
Release-notes.py used to depend on release already being tagged--now gathers release notes from previous tagged release up to current HEAD.
Closes#1857. Fixes bug where a transaction sending from a zaddr would have a priority of zero.
Transactions sent from a zaddr should now be mined sooner as they no longer have a priority of zero
Isolate verification to a ProofVerifier context object that allows verification behavior to be tuned by the caller.
This is an alternative foundation for #1892, i.e., #1892 will have to be changed if this PR is accepted.
I think this is a safer approach because it allows us to isolate verification behavior to a single object. This will come in handy when @arielgabizon finishes the batching code.
Fix anchor cache bug
Fixes#1912.
If an anchor is removed from the cache, but didn't exist in it beforehand, it will insert a blank tree. If it's reinserted in a child cache, when the child cache flushes it will mark the treestate as entered but won't bring the valid tree with it.
Thankfully, we assert when connecting blocks so that this inconsistency won't cause us to build on a blank tree after a reorg.
