feat: add BOM option to csv_agg

Support for BOM, requested on PostgREST/postgrest#1371 (comment).

```sql
select csv_agg(x, ',', true) from projects x;
      csv_agg
-------------------
id,name,client_id+
 1,Windows 7,1    +
 2,Windows 10,1   +
 3,IOS,2          +
 4,OSX,2          +
 5,Orphan,
(1 row)
```

Also possible to do with [media type handlers](https://docs.postgrest.org/en/v12/references/api/media_type_handlers.html#overriding-a-builtin-handler),
but it's more convenient to have it builtin.

Author: steve-chavez

.github/workflows/ci.yaml

@@ -37,7 +37,7 @@ jobs:
   loadtest:
     strategy:
       matrix:
-        kind: ['csv_agg', 'csv_agg_delim', 'postgrest']
+        kind: ['csv_agg', 'csv_agg_delim', 'csv_agg_delim_bom', 'postgrest']
     name: Loadtest
     runs-on: ubuntu-24.04
     steps:

README.md

@@ -35,6 +35,8 @@ select csv_agg(x) from projects x;
 (1 row)
 ```
 
+### Custom Delimiter
+
 It also supports adding a custom delimiter.
 
 ```psql
@@ -50,5 +52,22 @@ select csv_agg(x, '|') from projects x;
 (1 row)
 ```
 
-> [!IMPORTANT]
+> [!NOTE]
 > Newline, carriage return and double quotes are not supported as delimiters to maintain the integrity of the separated values format.
+
+### BOM
+
+You can include a byte-order mark (BOM) to make the CSV compatible with Excel.
+
+```psql
+select csv_agg(x, ',', true) from projects x;
+      csv_agg
+-------------------
+id,name,client_id+
+ 1,Windows 7,1    +
+ 2,Windows 10,1   +
+ 3,IOS,2          +
+ 4,OSX,2          +
+ 5,Orphan,
+(1 row)
+```

bench/csv_agg_delim_bom.sql

@@ -0,0 +1,5 @@
+\set lim random(1000, 2000)
+
+select csv_agg(t,',', true) from (
+  select * from student_emotion_assessments limit :lim
+) as t;

sql/pg_csv.sql

@@ -26,3 +26,15 @@ create aggregate csv_agg(anyelement) (
   finalfunc = csv_agg_finalfn,
   parallel  = safe
 );
+
+CREATE FUNCTION csv_agg_transfn(internal, anyelement, "char", boolean)
+  returns internal
+  language c
+  as 'pg_csv';
+
+CREATE AGGREGATE csv_agg(anyelement, "char", boolean) (
+  sfunc     = csv_agg_transfn,
+  stype     = internal,
+  finalfunc = csv_agg_finalfn,
+  parallel  = safe
+);

src/pg_csv.c

@@ -5,6 +5,7 @@ PG_MODULE_MAGIC;
 static const char NEWLINE = '\n';
 static const char DQUOTE  = '"';
 static const char CR      = '\r';
+static const char BOM[3]  = "\xEF\xBB\xBF";
 
 typedef struct {
   StringInfoData accum_buf;
@@ -50,9 +51,10 @@ static char *datum_to_cstring(Datum datum, Oid typeoid) {
 
 PG_FUNCTION_INFO_V1(csv_agg_transfn);
 Datum csv_agg_transfn(PG_FUNCTION_ARGS) {
-  CsvAggState    *state = !PG_ARGISNULL(0) ? (CsvAggState *)PG_GETARG_POINTER(0) : NULL;
-  HeapTupleHeader next  = !PG_ARGISNULL(1) ? PG_GETARG_HEAPTUPLEHEADER(1) : NULL;
-  char            delim = PG_NARGS() >= 3 && !PG_ARGISNULL(2) ? PG_GETARG_CHAR(2) : ',';
+  CsvAggState    *state    = !PG_ARGISNULL(0) ? (CsvAggState *)PG_GETARG_POINTER(0) : NULL;
+  HeapTupleHeader next     = !PG_ARGISNULL(1) ? PG_GETARG_HEAPTUPLEHEADER(1) : NULL;
+  char            delim    = PG_NARGS() >= 3 && !PG_ARGISNULL(2) ? PG_GETARG_CHAR(2) : ',';
+  bool            with_bom = PG_NARGS() >= 4 && !PG_ARGISNULL(3) ? PG_GETARG_BOOL(3) : false;
 
   // first call when the accumulator is NULL
   // pretty standard stuff, for example see the jsonb_agg transition function https://github.com/postgres/postgres/blob/3c4e26a62c31ebe296e3aedb13ac51a7a35103bd/src/backend/utils/adt/jsonb.c#L1521
@@ -80,6 +82,8 @@ Datum csv_agg_transfn(PG_FUNCTION_ARGS) {
   if (!state->header_done) {
     TupleDesc tdesc = lookup_rowtype_tupdesc(HeapTupleHeaderGetTypeId(next), HeapTupleHeaderGetTypMod(next));
 
+    if (with_bom) appendBinaryStringInfo(&state->accum_buf, BOM, sizeof(BOM));
+
     // build header row
     for (int i = 0; i < tdesc->natts; i++) {
       Form_pg_attribute att = TupleDescAttr(tdesc, i);

test/expected/bom.out

@@ -0,0 +1,21 @@
+-- this is done to avoid failing on a pure psql change that happened on postgres 16
+-- on pg <= 15 the BOM output adds one extra space, on pg 16 it doesn't
+\pset format unaligned
+\pset tuples_only on
+\echo
+
+-- include BOM (byte-order mark)
+SELECT csv_agg(x, ';', true) AS body
+FROM   projects x;
+id;name;client_id
+1;Windows 7;1
+2;has,comma;1
+;;
+4;OSX;2
+;"has""quote";
+5;"has,comma and ""quote""";7
+6;"has 
+ LF";7
+7;"has  CR";8
+8;"has 
+ CRLF""";8

test/sql/bom.sql

@@ -0,0 +1,9 @@
+-- this is done to avoid failing on a pure psql change that happened on postgres 16
+-- on pg <= 15 the BOM output adds one extra space, on pg 16 it doesn't
+\pset format unaligned
+\pset tuples_only on
+\echo
+
+-- include BOM (byte-order mark)
+SELECT csv_agg(x, ';', true) AS body
+FROM   projects x;