fix build on i386 without SSE2

Commit 3cba6b1 ("Use _Float16 for half conversions if available") added
support for using half-width float support in the compiler to perform
encoding operations, using the FLT16_MANT_DIG macro to check for
support on the given target.

However, on x86 GCC only supports this when SSE2 is enabled[1]. Unlike
clang and the other architectures where support for this is conditional,
GCC *does* define those macros even without SSE2 support, causing a
build failure:

  In file included from cborencoder_float.c:29:
  cborinternal_p.h: In function ‘encode_half’:
  cborinternal_p.h:56:5: error: invalid conversion to type ‘_Float16’ without option ‘-msse2’
     56 |     _Float16 f = (_Float16)x;
        |     ^~~~~~~~
  cborinternal_p.h: In function ‘decode_half’:
  cborinternal_p.h:65:5: error: invalid conversion from type ‘_Float16’ without option ‘-msse2’
     65 |     return (float)f;
        |

Work around this by additionally checking for this specific condition.

[1] https://gcc.gnu.org/onlinedocs/gcc/Half-Precision.html

Signed-off-by: Florian Larysch <fl@n621.de>

src/cborinternal_p.h

@@ -48,8 +48,12 @@
 /* Check for FLT16_MANT_DIG using integer comparison. Clang headers incorrectly
  * define this macro unconditionally when __STDC_WANT_IEC_60559_TYPES_EXT__
  * is defined (regardless of actual support for _Float16).
+ *
+ * GCC defines these macros but doesn't support arithmetic including
+ * conversions on x86 without SSE2.
  */
-#  if FLT16_MANT_DIG > 0 || __FLT16_MANT_DIG__ > 0
+#  if (FLT16_MANT_DIG > 0 || __FLT16_MANT_DIG__ > 0) && \
+      !(defined(__i386__) && !defined(__SSE2__))
 static inline unsigned short encode_half(float x)
 {
     unsigned short h;