sql.mk_select() calls: Removed no longer needed order_by=None when limit=0

sql.py: mk_select(): Set order_by to None if limit == 0

inputs/.TNRS/schema.sql: Documented that accepted names must be processed before any names that resolve to them, because the entry for the accepted name contains all the ranks parsed out but the resolved name of another entry contains just some ranks and the taxonomic name. Column-based import will do this automatically when the total # of rows is <= the partition_size (because _taxonconcept_set_matched_concept_id()'s accepted taxonconcept is created after the main taxonconcept), but TNRS has more rows than this so sorting is needed to ensure that all the accepted names are processed in the first partitions.

sql.py: table_order_by(): Cache the order_by in table.order_by and propagate it when a LIKE table is created

sql_gen.py: Table: Added order_by attr to cache the results of table_order_by()

sql.select() calls: Removed order_by=None everywhere that a stable row order is required (i.e. consistent between selects, or consistent between table transformations). This causes several tests to return different inserted row counts, because the input table is now being accessed in pkey order instead of in table order. This fixes a bug where tables with more rows than ~100 would return different results for repeated calls of the same non-ordered select.

sql.py: mk_select(): Use table_order_by() instead of table_pkey_col() to determine what column(s) to order by if order_by is set to order_by_pkey

sql.py: Added table_pkey_index(), index_order_by(), table_cluster_on(), table_order_by()

sql.py: Added index_exprs() and use it in index_cols()

README.TXT: Data import: On local machine: Added `make inputs/.TNRS/cleanup`, which is necessary because the PostgreSQL collation may differ between vegbiendev's and your DB