feat(griffin): added count(string), count(symbol) and count(long256). Also added rnd_long256(count) (#556)

Fixed "mask" values in all of hash tables. This was casing keys to be poorly distributed across arrays forcing high number of key collisions. I also created Long256HashSet and fixed a bug in SelectedRecord to do with addressing long256 column. All of that is tested.

This change deals with issues around protocol mechanics, in particular our responses to messages from the client on simple and extended mode.

For example:

- we now wait for sync message to send reply back, therefore grouping replies. This means that if client sends messages: P/B/D/E/S, we now send one reply back with 1/2/T/D/C/Z. This means that there is less chatter on the wire because 1. we are sending less messages and 2. the client does not get confused and therefore send less messages.
- for each message sent by client, we send an acknowledgement back
- stop sending spurious messages
- named prepared statements are cached on parse and we correctly bind parameters when receiving named statement on the Bind and Describe messages
- when executing 'select', we now send back the number of returned rows
- when executing 'set', we now send correct query tag (SET)
- for empty queries, we reply back with correct message type
- for queries with no data, we reply back with 'no data' message type
- row description now includes column number
- row descriptions attempts to populate type size and type modifier

fix(griffin): fixed issues with long NULL converting to double NULL. Also fixed group-by validation logic to deal with clauses more systematically. Fixed #540 (#545)

A new implementation of virtual memory that resizes a single page rather than using a list of pages that do no resize. This implementation includes support for memory mapped page resize.

Overall single memory page method improves performance for both read and write, also making data framing much easier

I also fixed a problem, raised by users, where uploading "bad" timestamp in CSV would result in exception and perceived curl hang (no response from server)

fix(griffin): fixed 'with' clause behaviour in set operations (union, expct, intersect), in recursive 'with' and in sub-queries (#541)

I had to move telemetry queue and sequences to CairoEngine. Leaving them on MessageBus would cause problem when various components run with local pools. In this scenario there are multiple copies of MessageBus and it would be unnacceptable to have multiple queues for telemetry. Only one queue would be consumed and all others stalling.

Additionally requiring PropServerConfiguration for very simple embedded setup was unreasonable. Telemetry should be ok with CairoConfiguration.

I moved MessageBus instance inside CairoEngine. They are created based on the same configuration now and they are used in 1-to-1 combination most of the time. Except of course for those pesky local pools. The components that can have local thread pool use separate instance of MessageBus.

Functions like first(symbol) could not work in any of the aggregation cursors. Access to symbol table was complex and incorrect. Functions were initialized with wrong cursor and symbol table access was overcomplicated to compensate for wrong cursor!

windows does not support wildcards in file names (2020, eh?) so the release version is hardcoded

fix(griffin): maven test runner (surefire) was picking up default encoding and thus failing tests (#514)

Also removed duplicate operator expression.