Fix reading UTF-8 encoded sample names when char is signed

The trick used in bcf_hdr_parse_sample_line() to rapidly find tabs
and newlines could be defeated by UTF-8 characters outside the
Basic Latin range on platforms where "char" is signed (like x86).
It's currently not clear if VCF intends to allow these, but the
4.3 specification does allow UTF-8 and it's easy enough to support.
Fix by casting to unsigned when making the comparison.

Modifies formatcols.vcf to include a UTF-8 character for a
round-trip test.

Fixes samtools/bcftools#1408

test/formatcols.vcf

@@ -2,5 +2,5 @@
 ##FILTER=<ID=PASS,Description="All filters passed">
 ##contig=<ID=1>
 ##FORMAT=<ID=S,Number=1,Type=String,Description="Text">
-#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	S1	S2	S3
+#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	S1	S²	S3
 1	100	a	A	T	.	.	.	S	a	bbbbbbb	ccccccccc

vcf.c

@@ -150,7 +150,7 @@ int HTS_RESULT_USED bcf_hdr_parse_sample_line(bcf_hdr_t *h, const char *str)
     const char *p, *q;
     // add samples
     for (p = q = str;; ++q) {
-        if (*q > '\n') continue;
+        if ((unsigned char) *q > '\n') continue;
         if (++i > 9) {
             if ( bcf_hdr_add_sample_len(h, p, q - p) < 0 ) ret = -1;
         }