Custom cursors support

[bitgorbovsky]

There is a problem with cursor based pagination on requests by secondary indexes. Consider a simple key-value schema:

field	type	info
key	string	primary key
value	string
tag	string	secondary index, non unique

The call crud.select('kv', {{'=', 'tag', 'tagValue'}}, {first = 10, after = ...}) makes a fullscan of secondary index from starting point (tagValue) to tuple equaling cursor. Under heavy load it leads to CPU consumption and cluster performance degradation.

To solve this problem, we have to create a special unique index paging with two fields {tag, key} and implement pagination manually, by executing such calls like

local rs = crud.select('kv', {{'==', 'paging', {'tagValue', box.NULL}}}, {first = 10})

...

local last_key = rs.ros[#rs.rows][0]
rs = crud.select('kv', {{'>', 'paging', {'tagValue', last_key}}}, {first = 10})
...

It would be nice to introduce some feature like passing name of paging index and using key from this index as cursors

[DifferentialOrange]

We also may consider supporting new pagiation here: tarantool/tarantool#7639