Fixes BCF in-memory decoding with 64-bit integers.

Any 64-bit INFO field that wasn't the last in the list would cause
subsequent fields to be decoded incorrectly.

This commit fixes that, plus updates the tests accordingly so the bug
could be triggered.

Fixes the first part of #999 (test1.vcf), but doesn't fix the second
part (BCF output silently being broken).

Fixes samtools/bcftools#1123

test/longrefs/index.expected2.vcf

@@ -1 +1 @@
-1	10010000110	.	G	<DEL>	0	.	SVTYPE=DEL;SVLEN=-890;END=10010001000	PL	0,1,45
+1	10010000110	.	G	<DEL>	0	.	SVTYPE=DEL;SVLEN=-890;END=10010001000;QS=1,0	PL	0,1,45

test/longrefs/index.vcf

@@ -213,4 +213,4 @@
 1	10010000107	.	G	<*>	0	.	DP=1;I16=0,1,0,0,33,1089,0,0,29,841,0,0,2,4,0,0;QS=1,0;MQ0F=0	PL	0,3,29
 1	10010000108	.	C	<*>	0	.	DP=1;I16=0,1,0,0,32,1024,0,0,29,841,0,0,1,1,0,0;QS=1,0;MQ0F=0	PL	0,3,29
 1	10010000109	.	A	<*>	0	.	DP=1;I16=0,1,0,0,35,1225,0,0,29,841,0,0,0,0,0,0;QS=1,0;MQ0F=0	PL	0,3,29
-1	10010000110	.	G	<DEL>	0	.	SVTYPE=DEL;SVLEN=-890;END=10010001000	PL	0,1,45
+1	10010000110	.	G	<DEL>	0	.	SVTYPE=DEL;SVLEN=-890;END=10010001000;QS=1,0	PL	0,1,45

vcf.c

@@ -59,7 +59,7 @@ HTSLIB_EXPORT
 uint32_t bcf_float_vector_end = 0x7F800002;
 
 HTSLIB_EXPORT
-uint8_t bcf_type_shift[] = { 0, 0, 1, 2, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+uint8_t bcf_type_shift[] = { 0, 0, 1, 2, 3, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
 
 static bcf_idinfo_t bcf_idinfo_def = { .info = { 15, 15, 15 }, .hrec = { NULL, NULL, NULL}, .id = -1 };