seq: revert Solaris 8 work-around that caused x86 regression

* src/seq.c: Don't include <math.h>, <float.h>.
(abs_rel_diff): Remove.
(print_numbers): Test for equality, not for an epsilonish value.
This reverts 4827dd27b0, aka
v6.10-185-g4827dd2, which broke 'seq' on the x86; for example, it
causes "seq 9223372036854775807 9223372036854775808" to incorrectly
output 3 numbers instead of 2.  It's better to punish obsolescent
hosts that have incorrectly-working floating-point than to punish
correctly-working hosts.
* tests/misc/seq: Use 0.9000000000000, rather than
0.90000000000000000000, to avoid tickling a bug in Solaris 8 strtold,
which converts "0.9" and "0.9000000000000" correctly, but incorrectly
converts "0.90000000000000000000" to a smaller value.

tests/misc/seq

@@ -47,7 +47,10 @@ my @Tests =
     {OUT_SUBST => 's,^-0\.0$,0.0,'},
    ],
    ['float-5',	qw(0.8 1e-1 0.9),	{OUT => [qw(0.8 0.9)]}],
-   ['float-6',	qw(0.8 0.1 0.90000000000000000000),	{OUT => [qw(0.8 0.9)]}],
+   # Don't append lots of zeros to that 0.9000...; for example, changing the
+   # number to 0.90000000000000000000 tickles a bug in Solaris 8 strtold
+   # that would cause the test to fail.
+   ['float-6',	qw(0.8 0.1 0.9000000000000),	{OUT => [qw(0.8 0.9)]}],
 
    ['wid-1',	qw(.8 1e-2 .81),  {OUT => [qw(0.80 0.81)]}],
    ['wid-2',	qw(.89999 1e-7 .8999901),  {OUT => [qw(0.8999900 0.8999901)]}],