perf(parser): co-allocate small slices and size lists exactly [W4-04]

CHANGELOG.md

@@ -44,6 +44,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
     are removed.
 
 ### Performance
+- **Parser**: function arguments, the SELECT list, GROUP BY and ORDER BY are now collected on parser-owned scratch stacks and copied once into exact-size slices (previously grown with `append` from a fixed capacity). A `FunctionCall` with up to three arguments shares one allocation with its argument array, and a `SelectStatement` shares one with its single `From` entry and, for up to four columns, its `Columns` array. Parse allocations drop 28% on `aggregate` (18 to 13) and 12% on `complex` (50 to 44), and instructions per parse drop 1-5% on every SELECT shape. Time on the amd64 CI gate (five runs on three runner CPU models): `simple_select` 4-5% faster, `subquery` 3-6% faster; the other SELECT shapes vary by runner from about 4% faster to 4% slower. INSERT, UPDATE and DELETE execute the same instructions as before but measured up to 5% slower on AMD Zen 5 runners (`insert` +5% in two of three runs, and the mixed `ParseRW/sqlite/medium` corpus up to +5%), and within 2% on the other runner CPUs. Zero-instruction-change controls (only padding `Parser` or only moving code in the binary) moved single benchmarks by +3.4% to -4.5% on the same gate, so these are most likely code and data placement effects rather than added work. User-visible: `Columns`, `GroupBy`, `OrderBy` and `FunctionCall.Arguments` now have `len == cap` (they used to carry spare capacity), so a caller that appends to them reallocates instead of writing into spare space; trees are otherwise identical (guarded by the behaviour goldens). No exported API change. [W4-04]
 - **`ast.ReleaseAST` is O(1) for parser-built trees that hold no pooled nodes**: the parser counts the nodes it draws from the AST pools and tells
   `ReleaseAST` through an internal hook (`pkg/sql/internal/astrelease`). When
   the count is zero, which covers ordinary SELECT/INSERT/UPDATE/DELETE

pkg/gosqlx/alloc_budget_test.go

@@ -39,14 +39,14 @@ import (
 
 // parseAllocBudget is allocs/op of Parse+ReleaseAST for the 11 portable queries.
 var parseAllocBudget = map[string]float64{
-	"simple_select": 5,
-	"select_where":  18,
-	"join":          17,
-	"aggregate":     18,
-	"subquery":      14,
-	"cte":           21,
-	"window":        11,
-	"complex":       50,
+	"simple_select": 3,
+	"select_where":  16,
+	"join":          15,
+	"aggregate":     13,
+	"subquery":      10,
+	"cte":           16,
+	"window":        9,
+	"complex":       44,
 	"insert":        17,
 	"update":        13,
 	"delete":        5,
@@ -69,14 +69,14 @@ var dialectAllocStatement = map[string]string{
 // dialectAllocBudget is allocs/op of ParseWithDialect+ReleaseAST for the
 // statement pinned in dialectAllocStatement.
 var dialectAllocBudget = map[string]float64{
-	"postgresql": 37,
+	"postgresql": 29,
 	"mysql":      50,
 	"mariadb":    61,
-	"sqlserver":  36,
-	"oracle":     68,
-	"sqlite":     48,
-	"snowflake":  71,
-	"clickhouse": 82,
+	"sqlserver":  28,
+	"oracle":     52,
+	"sqlite":     47,
+	"snowflake":  53,
+	"clickhouse": 80,
 }
 
 const allocRuns = 200