Default sparse read via tiledb_array to UNORDERED

[eddelbuettel]

This PR changes the default query layout for read access via tiledb_array() and the [ accessor to 'UNORDERED' for sparse matrices (and 'ROW_MAJOR' for dense). The read setting now matches the [<- write access implementation.

Unordered generally performs better, motivating the change. This is a change in behavior as seen by the handful of tests which needed updating.

As an illustration, here is a approximately seven-fold gain with a (local disk) array using the Deutsche Boerse (csv to sparse array) data set from a tutorial from last year:

$ ./exampleDBoerse.R        
                 test replications elapsed relative                                                      
1      a <- resNone[]           10  10.663    7.451                                                      
2 b <- resUnordered[]           10   1.431    1.000                                                      
$ 

[shortcut-integration]

This pull request has been linked to Shortcut Story #23685: Performance issue running R query..