Improve floating point equality test

This allows for a relative error rather than an absolute (which would set
an artificial scale). This basically says “we trust our data (which comes
from the dive computer’s measurement after all) to a certain number of
significant digits” rather than “we will never encounter anything smaller
than 1 / a million but not zero” which would be awfully unit dependent.

Signed-off-by: Robert C. Helling <helling@atdotde.de>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>

dive.h

@@ -20,7 +20,7 @@
     (void) (&_max1 == &_max2);      \
     _max1 > _max2 ? _max1 : _max2; })
 
-#define IS_FP_SAME(_a,_b) (abs((_a) - (_b)) < 0.000001)
+#define IS_FP_SAME(_a,_b) (abs((_a) - (_b)) < 0.000001 * MAX(abs(_a), abs(_b)))
 
 #include <libxml/tree.h>
 #include <libxslt/transform.h>