workload/schemachange: handle crdb_internal computed cols in colIsRefByComputed

[annrpom]

Our query for colIsRefByComputed cannot handle the following case (human-readable ops generated by schemachange workload):

[email protected]:26257/demoapp/movr> CREATE TABLE public.table_w0_1 (                                        
                                ->     "co\\u58A1l1 _w0_8" INTERVAL NOT NULL,                              
                                ->     INDEX (                                                             
                                ->         (CASE                                                           
                                ->             WHEN "co\\u58A1l1 _w0_8" IS NULL THEN ')':::STRING          
                                ->             ELSE e'-s\\x12M\\bps':::STRING                              
                                ->         END) DESC                                                       
                                ->     )                                                                   
                                -> );                                                                      
CREATE TABLE

Time: 10ms total (execution 10ms / network 0ms)

[email protected]:26257/demoapp/movr> ALTER TABLE public.table_w0_1                                           
                                -> DROP COLUMN "co\\u58A1l1 _w0_8";                                        
ERROR: column "co\\\\u58A1l1 _w0_8" is referenced by computed column "crdb_internal_idx_expr"
SQLSTATE: 42P10

Jira issue: CRDB-40208

Epic CRDB-19168

[annrpom]

From Faizan:

I think this would be similar to: getUniqueConstraintsForTable and the whole unique constraint logic that detects column references.

[rafiss]

Let's fix this as part of enabling DROP COLUMN (#127286):

cockroach/pkg/workload/schemachange/operation_generator.go

Lines 1631 to 1635 in e77f714

	// TODO(#126967): We need to add a check for the column being in an expression
	// to an index. In the case where the expression does not already exist for
	// us to use, we add an internal crdb_internal_idx_expr prefixed column to
	// the table.
	stmt.potentialExecErrors.add(pgcode.InvalidColumnReference)