fmt: slim down binary size of Temporal ISO 8601 duration printer

[BurntSushi]

In a simple benchmark program that prints spans, signed and unsigned
durations using jiff::fmt::temporal::SpanPrinter, the total number of
LLVM lines is reduced by about 6,000.

There is also a modest improvement in runtime performance:

$ critcmp baseline x01 -f 'iso8601.*jiff'
group                                                       baseline                               x05
-----                                                       --------                               ---
print/display/duration/iso8601/jiff                         1.61     27.4±0.24ns        ? ?/sec    1.00     17.1±0.18ns        ? ?/sec
print/display/span/iso8601/jiff                             1.79     30.6±0.36ns        ? ?/sec    1.00     17.1±0.09ns        ? ?/sec
print/iso8601_duration/long-time/duration/buffer/jiff       1.26     24.3±0.07ns        ? ?/sec    1.00     19.3±0.07ns        ? ?/sec
print/iso8601_duration/long-time/duration/to_string/jiff    2.08     52.9±0.28ns        ? ?/sec    1.00     25.4±0.08ns        ? ?/sec
print/iso8601_duration/long-time/span/buffer/jiff           1.39     29.2±0.09ns        ? ?/sec    1.00     21.0±0.04ns        ? ?/sec
print/iso8601_duration/long-time/span/to_string/jiff        2.14     59.0±0.55ns        ? ?/sec    1.00     27.6±0.07ns        ? ?/sec
print/iso8601_duration/long/span/buffer/jiff                1.76     28.1±0.08ns        ? ?/sec    1.00     16.0±0.08ns        ? ?/sec
print/iso8601_duration/long/span/to_string/jiff             1.84     40.1±0.15ns        ? ?/sec    1.00     21.8±0.06ns        ? ?/sec
print/iso8601_duration/medium/span/buffer/jiff              1.37     17.4±0.07ns        ? ?/sec    1.00     12.8±0.08ns        ? ?/sec
print/iso8601_duration/medium/span/to_string/jiff           1.65     31.6±0.14ns        ? ?/sec    1.00     19.2±0.08ns        ? ?/sec
print/iso8601_duration/short/duration/buffer/jiff           1.18     14.7±0.03ns        ? ?/sec    1.00     12.4±0.04ns        ? ?/sec
print/iso8601_duration/short/duration/to_string/jiff        1.08     22.3±0.11ns        ? ?/sec    1.00     20.6±0.08ns        ? ?/sec
print/iso8601_duration/short/span/buffer/jiff               1.27     14.9±0.06ns        ? ?/sec    1.00     11.8±0.02ns        ? ?/sec
print/iso8601_duration/short/span/to_string/jiff            1.14     22.7±0.15ns        ? ?/sec    1.00     19.9±0.10ns        ? ?/sec
print/iso8601_duration/tiny/duration/buffer/jiff            1.00      9.2±0.03ns        ? ?/sec    1.26     11.6±0.01ns        ? ?/sec
print/iso8601_duration/tiny/duration/to_string/jiff         1.00     12.0±0.05ns        ? ?/sec    1.19     14.3±0.10ns        ? ?/sec
print/iso8601_duration/tiny/span/buffer/jiff                1.04     11.4±0.11ns        ? ?/sec    1.00     11.0±0.06ns        ? ?/sec
print/iso8601_duration/tiny/span/to_string/jiff             1.01     14.1±0.08ns        ? ?/sec    1.00     14.0±0.10ns        ? ?/sec

The only regression here is on a very tiny duration (PT2S). I'm
able to eliminate most of the regression, but only by inlining the
print_unsigned_duration_impl function, which increases code size by a
few hundred LLVM lines. My feeling at present is that this isn't worth
it.

The runtime improvements widen the gap between Jiff and Chrono. Before:

$ critcmp baseline -g '(.*/)(?:jiff|chrono)$' -f 'iso8601.*buffer'
group                                                baseline/chrono                        baseline/jiff
-----                                                ---------------                        -------------
print/iso8601_duration/long-time/duration/buffer/    1.91     46.3±0.32ns        ? ?/sec    1.00     24.3±0.07ns        ? ?/sec
print/iso8601_duration/long-time/span/buffer/                                               1.00     29.2±0.09ns        ? ?/sec
print/iso8601_duration/long/span/buffer/                                                    1.00     28.1±0.08ns        ? ?/sec
print/iso8601_duration/medium/span/buffer/                                                  1.00     17.4±0.07ns        ? ?/sec
print/iso8601_duration/short/duration/buffer/        2.19     32.1±0.29ns        ? ?/sec    1.00     14.7±0.03ns        ? ?/sec
print/iso8601_duration/short/span/buffer/                                                   1.00     14.9±0.06ns        ? ?/sec
print/iso8601_duration/tiny/duration/buffer/         3.39     31.1±0.18ns        ? ?/sec    1.00      9.2±0.03ns        ? ?/sec
print/iso8601_duration/tiny/span/buffer/                                                    1.00     11.4±0.11ns        ? ?/sec

After:

$ critcmp cmp -g '(.*/)(?:jiff|chrono)$' -f 'iso8601.*buffer'
group                                                cmp/chrono                             cmp/jiff
-----                                                ----------                             --------
print/iso8601_duration/long-time/duration/buffer/    2.41     46.5±0.23ns        ? ?/sec    1.00     19.3±0.03ns        ? ?/sec
print/iso8601_duration/long-time/span/buffer/                                               1.00     21.2±0.04ns        ? ?/sec
print/iso8601_duration/long/span/buffer/                                                    1.00     16.2±0.05ns        ? ?/sec
print/iso8601_duration/medium/span/buffer/                                                  1.00     12.7±0.03ns        ? ?/sec
print/iso8601_duration/short/duration/buffer/        2.60     31.7±0.15ns        ? ?/sec    1.00     12.2±0.04ns        ? ?/sec
print/iso8601_duration/short/span/buffer/                                                   1.00     11.6±0.04ns        ? ?/sec
print/iso8601_duration/tiny/duration/buffer/         4.09     31.3±0.12ns        ? ?/sec    1.00      7.7±0.07ns        ? ?/sec
print/iso8601_duration/tiny/span/buffer/                                                    1.00      7.1±0.05ns        ? ?/sec