fix(events): order events consistently when querying

[rvagg]

(This PR is against #12421 until it's merged)

In ChainIndexer and the existing master version of this same code, we do an ORDER BY when querying for events but then we do a slices.Sort on the output, only sorting by epoch.

1. The sort is not stable, so it can end up with events jumbled up even if the epoch is properly sorted
2. We're not making use of the ORDER BY, in fact we do the reverse of what we ask for, so we should just do it there
3. The existing ORDER BY relies on insertion order (_rowid_) of the event entries, which should mostly be fine (aside from reverts perhaps), but we have event_index that we use for logIndex on Eth output and we're not making use of that or properly ordering by it.

Here's a list of eth events from epoch 4374648, just listing their blockNumber, transactionIndex and logIndex:

  [ '0x42c078', '0x49', '0x80' ],
  [ '0x42c078', '0x35', '0x62' ],
  [ '0x42c078', '0x35', '0x61' ],
  [ '0x42c078', '0x35', '0x60' ],
  [ '0x42c078', '0x49', '0x81' ],
  [ '0x42c078', '0x49', '0x82' ]

They should be nicely ordered. Here is the same but with this fix:

  [ '0x42c078', '0x35', '0x60' ],
  [ '0x42c078', '0x35', '0x61' ],
  [ '0x42c078', '0x35', '0x62' ],
  [ '0x42c078', '0x49', '0x80' ],
  [ '0x42c078', '0x49', '0x81' ],
  [ '0x42c078', '0x49', '0x82' ]

Note that in the process here we're going with ASC for all of the fields - epochs, messages, events and event entries.

This is the query now for a from/to query of eth_getLogs:

SELECT e.id,
  tm.height, tm.tipset_key_cid,
  e.emitter_id, e.emitter_addr,
  e.event_index, tm.message_cid,
  tm.message_index, e.reverted,
  ee.flags, ee.key,
  ee.codec, ee.value
FROM event e  JOIN tipset_message tm ON e.message_id = tm.id
JOIN event_entry ee ON e.id = ee.event_id
HERE tm.height BETWEEN ? AND ? AND e.reverted=?
ORDER BY tm.height ASC, tm.message_index ASC, e.event_index ASC, ee._rowid_ ASC

And here's what the query plan looks like for that:

QUERY PLAN
|--SEARCH tm USING INDEX idx_height (height>? AND height<?)
|--SEARCH e USING INDEX idx_event_message_id (message_id=?)
|--SEARCH ee USING INDEX event_entry_event_id (event_id=?)
`--USE TEMP B-TREE FOR RIGHT PART OF ORDER BY

As long as we're hitting the height index first then we should be good.

Here's a larger range of epochs to see that it's doing ordering across epochs:

  [ '0x42c0db', '0x35', '0x78' ],
  [ '0x42c0de', '0x3c', '0x7c' ],
  [ '0x42c0e1', '0x37', '0x61' ],
  [ '0x42c0e4', '0x37', '0x63' ],
  [ '0x42c0e4', '0x39', '0x65' ],
  [ '0x42c0e7', '0x39', '0x6e' ],
  [ '0x42c0e9', '0x54', '0x9f' ],
  [ '0x42c0ec', '0x3b', '0x64' ],
  [ '0x42c0ee', '0x29', '0x58' ],
  [ '0x42c0ee', '0x29', '0x59' ],
  [ '0x42c0ee', '0x29', '0x5a' ],
  [ '0x42c0ee', '0x29', '0x5b' ],
  [ '0x42c0ee', '0x29', '0x5c' ],
  [ '0x42c0ee', '0x29', '0x5d' ],
  [ '0x42c0ee', '0x29', '0x5e' ],
  [ '0x42c0ee', '0x29', '0x5f' ]

[aarshkshah1992]

@rvagg This is a nice UX improvement and we should merge it.