SQL: improve lazy execution threading model

Currently "lazy" mode creates new thread for every query. Moreover, messages are queued to query pool first and then re-submitted to target thread. This makes execution of small queries inefficient.

We need to do the following:
1) Define separate pool for lazy execution
2) It's size should be about 4 x cores, due to blocking nature of lazy execution
3) Set thread timeouts
4) Submit new query execute request to pool's shared queue from GridIoManager directly
5) Submit "next page" requests to directly to proper thread, using [nodeId, queryId, segment] triplet
6) Cancel request can be submitted to query pool and re-submitted to appropriate threads afterwards.