Add whole-record WAL compression alongside FPI compression

When a record is larger than wal_compression_threshold, compress it as a
single unit instead of compressing each full-page image separately.  This
wins whenever the images in one record share content, which is typical of
B-tree index builds: CREATE INDEX over 10M random floats drops from 160MB
of WAL to 132MB with lz4, and from 125MB to 97MB with zstd.

Only lz4 and zstd take part.  pglz refuses input it cannot shrink by 25%,
which a record made of full-page images rarely clears, so it saved
nothing while costing an extra compression attempt.  It still compresses
full-page images as before.

Assembly skips per-FPI compression while whole-record compression is
attempted, and reassembles the record if that does not pay off.  Setting
wal_compression_threshold above the largest possible record restores the
previous behaviour.

The per-block compressed_page arrays make way for two buffers of the same
maximum size, one staging the record and one taking the compressor
output.  XLR_COMPRESSED in xl_info marks a compressed record, which
begins with an XLogCompressionHeader giving the method and the
decompressed length.

Author: Andrey Borodin
Discussion: https://postgr.es/m/4DC38068-976E-4A84-8EE6-4EFACBBD927A@yandex-team.ru

Author: x4m

src/backend/access/transam/xlog.c

@@ -142,6 +142,7 @@ int			wal_retrieve_retry_interval = 5000;
 int			max_slot_wal_keep_size_mb = -1;
 int			wal_decode_buffer_size = 512 * 1024;
 bool		track_wal_io_timing = false;
+int			wal_compression_threshold = 512;
 
 #ifdef WAL_DEBUG
 bool		XLOG_DEBUG = false;
@@ -751,6 +752,22 @@ static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
 
 static void XLogChecksums(uint32 new_type);
 
+#ifdef WAL_DEBUG
+/* Read length of a record, accounting for possible compression */
+static uint32
+XLogGetRecordTotalLen(XLogRecord *record)
+{
+	if (record->xl_info & XLR_COMPRESSED)
+	{
+		XLogCompressionHeader *c = (XLogCompressionHeader *) record;
+
+		Assert(c->decompressed_length > 0);
+		return c->decompressed_length;
+	}
+	return record->xl_tot_len;
+}
+#endif
+
 /*
  * Insert an XLOG record represented by an already-constructed chain of data
  * chunks.  This is a low-level routine; to construct the WAL record header
@@ -1068,7 +1085,7 @@ XLogInsertRecord(XLogRecData *rdata,
 		/* We also need temporary space to decode the record. */
 		record = (XLogRecord *) recordBuf.data;
 		decoded = (DecodedXLogRecord *)
-			palloc(DecodeXLogRecordRequiredSpace(record->xl_tot_len));
+			palloc(DecodeXLogRecordRequiredSpace(XLogGetRecordTotalLen(record)));
 
 		if (!debug_reader)
 			debug_reader = XLogReaderAllocate(wal_segment_size, NULL,