Rework storage layer - Githubissues

Goal of this PR

Fixes https://github.com/optakt/flow-dps/issues/522

This PR, along with https://github.com/optakt/golang-lru/pull/1 redesigns the storage system of the trie improvements:

Instead of being a call back, the function given to the LRU cache is an eviction handler which is called BEFORE an eviction happens
If this handler returns false, the eviction is aborted, and the element that was supposed to get evicted gets pushed back to the front of the eviction list

In the store, we run the following logic in our eviction handler:

    s.cache, err = lru.NewWithEvict(config.CacheSize, func(k, v interface{}, used int) bool {
        hash, ok := k.(hash.Hash)
        if !ok {
            logger.Fatal().Interface("got", k).Msg("unexpected key format")
        }

        // If the entry is dirty and the cache is not full, abort eviction.
        s.isDirtyMu.RLock()
        if s.isDirty[hash] && used < s.cacheSize {
            s.isDirtyMu.RUnlock()
            return false
        }
        s.isDirtyMu.RUnlock()

        // If the cache is full of dirty entries, push the current commit to disk and evict the entry.
        if s.cacheFullAndDirty(used) {
            s.forceCommit()
        }

        // If the current entry is clean, we can evict it. Otherwise, abort the eviction.
        return !s.isDirty[hash]
    })

If the entry that should be evited is dirty and there is still space in the cache, we can safely abort the eviction. Aborting it will result in this entry going back all the way to the other side of the eviction list.
If the cache is full and that all of its entries are dirty, we force the current transaction to be committed and evict the entry.
Otherwise, determine whether or not to evict the value based on whether it is dirty.

Note: This PR includes changes that add error returns to the Insert method, which will conflict with the changes that @awfm9 will bring later on. It should be easy to solve those conflicts however: we just take the code from Max and add error handling in the only remaining case where we create a leaf.

optakt / flow-dps

Rework storage layer #523

Goal of this PR

Checklist

Does the PR require tests to be added or updated?

Does the PR require documentation to be added or updated?

Misc