prototype skip encoding

[max-hoffman]

prototype before/after:

BenchmarkOltpIndexScan-12    	     206	   5190640 ns/op	 5243491 B/op	  140229 allocs/op
BenchmarkOltpIndexScan-12    	     650	   1785686 ns/op	 3883499 B/op	   20767 allocs/op

The premise of this is a custom row type:

type ProllyRow struct {
	key, value val.Tuple
	kd, vd     val.TupleDesc
	ords       []int // placement access
}

func (r ProllyRow) GetBytes(i int, typ sql.Type) ([]byte, error) {
	pos := r.ords[i]

	if pos < r.kd.Count() {
		ret := r.kd.GetField(pos, r.key)
		return ret, nil
	}

	pos -= r.kd.Count()
	ret := r.vd.GetField(pos, r.value)
	return ret, nil
}

The idea is to go from val.Tuple straight to wire values:

	outVals := make([]sqltypes.Value, len(sch))
	var err error
	if br, ok := row.(sql.BytesRow); ok {
		for i, col := range sch {
			buf, err := br.GetBytes(i, col.Type)
			outVals[i] = sqltypes.MakeTrusted(col.Type.Type(), buf)
			if err != nil {
				return nil, err
			}
		}
		return outVals, nil
	}

• correctness considerations for when storage encoding is not consistent with MySQL byte encoding
• certain values will have to be materialized to satisfy engine logic that operates on Go types
• the ordinal mapping above is 1-step, while main has 2-steps
• this doesn't work with virtual columns

[max-hoffman]

#benchmark

[github-actions]

@max-hoffman workflow run: https://github.com/dolthub/dolt/actions/runs/12126600902

[max-hoffman]

#benchmark

[github-actions]

@max-hoffman workflow run: https://github.com/dolthub/dolt/actions/runs/12129076658

[coffeegoddd]

@max-hoffman DOLT

read_tests	from_latency	to_latency	percent_change
covering_index_scan	0.62	0.62	0.0
groupby_scan	16.71	17.95	7.42
index_join	2.26	2.39	5.75
index_join_scan	1.44	1.52	5.56
index_scan	55.82	14.46	-74.1
oltp_point_select	0.27	0.27	0.0
oltp_read_only	5.47	5.47	0.0
select_random_points	0.67	0.64	-4.48
select_random_ranges	0.65	0.67	3.08
table_scan	55.82	14.46	-74.1
types_table_scan	147.61	39.65	-73.14

write_tests	from_latency	to_latency	percent_change
oltp_delete_insert	6.21	6.21	0.0
oltp_insert	3.07	3.07	0.0
oltp_read_write	11.65	10.84	-6.95
oltp_update_index	3.13	0.45	-85.62
oltp_update_non_index	3.07	0.45	-85.34
oltp_write_only	6.21	5.28	-14.98
types_delete_insert	6.55	6.67	1.83

[max-hoffman]

#benchmark

[github-actions]

@max-hoffman workflow run: https://github.com/dolthub/dolt/actions/runs/12131882450

[coffeegoddd]

@max-hoffman DOLT

read_tests	from_latency	to_latency	percent_change
covering_index_scan	0.61	0.62	1.64
groupby_scan	16.41	17.63	7.43
index_join	2.26	2.39	5.75
index_join_scan	1.44	1.52	5.56
index_scan	54.83	15.0	-72.64
oltp_point_select	0.26	0.26	0.0
oltp_read_only	5.28	5.28	0.0
select_random_points	0.63	0.62	-1.59
select_random_ranges	0.63	0.64	1.59
table_scan	55.82	14.46	-74.1
types_table_scan	142.39	39.65	-72.15

write_tests	from_latency	to_latency	percent_change
oltp_delete_insert	6.21	6.21	0.0
oltp_insert	3.07	3.07	0.0
oltp_read_write	11.45	11.45	0.0
oltp_update_index	3.13	3.13	0.0
oltp_update_non_index	3.02	3.02	0.0
oltp_write_only	6.21	6.21	0.0
types_delete_insert	6.55	6.55	0.0