Avoid contention on Instant::now() calls

[loyd]

• Instant::duration_since now saturates to zero.
• Add the contention benchmark and store results in benches/README.md.
• Instant::elapsed is added to make Instant drop-in replacement of std::time::Instant.
• Add generation of links to diffs in CHANGELOG.md (according to keepachangelog spec).

Results (performed on an AMD Ryzen 7 4800HS CPU):

group            main                                    no-contention
-----            ----                                    -------------
quanta/now/1     1.33     18.8±9.87ns        ? ?/sec     1.00     14.2±7.40ns        ? ?/sec
quanta/now/10    13.38  345.7±54.28ns        ? ?/sec     1.00    25.8±10.19ns        ? ?/sec
quanta/now/11    15.72  394.6±37.46ns        ? ?/sec     1.00     25.1±8.89ns        ? ?/sec
quanta/now/12    17.74  457.2±59.94ns        ? ?/sec     1.00     25.8±9.02ns        ? ?/sec
quanta/now/2     3.23    53.6±12.42ns        ? ?/sec     1.00     16.6±5.66ns        ? ?/sec
quanta/now/3     6.11   100.7±35.62ns        ? ?/sec     1.00     16.5±5.76ns        ? ?/sec
quanta/now/4     9.31   160.2±43.22ns        ? ?/sec     1.00     17.2±5.69ns        ? ?/sec
quanta/now/5     10.32  182.0±38.35ns        ? ?/sec     1.00     17.6±5.71ns        ? ?/sec
quanta/now/6     11.92  205.4±61.60ns        ? ?/sec     1.00     17.2±5.86ns        ? ?/sec
quanta/now/7     13.92  244.1±65.22ns        ? ?/sec     1.00     17.5±6.55ns        ? ?/sec
quanta/now/8     15.56  269.7±60.34ns        ? ?/sec     1.00     17.3±6.17ns        ? ?/sec
quanta/now/9     12.79  309.2±87.00ns        ? ?/sec     1.00     24.2±9.47ns        ? ?/sec
stdlib/now/1     1.03  1165.6±142.16ns        ? ?/sec    1.00  1126.9±71.75ns        ? ?/sec
stdlib/now/10    1.00  1368.3±30.03ns        ? ?/sec     1.01  1386.1±75.49ns        ? ?/sec
stdlib/now/11    1.00  1401.8±70.33ns        ? ?/sec     1.00  1401.1±68.81ns        ? ?/sec
stdlib/now/12    1.00  1389.6±48.94ns        ? ?/sec     1.00  1387.9±38.73ns        ? ?/sec
stdlib/now/2     1.00  1358.9±105.26ns        ? ?/sec    1.05  1431.1±147.48ns        ? ?/sec
stdlib/now/3     1.00  1380.8±55.35ns        ? ?/sec     1.00  1382.1±73.67ns        ? ?/sec
stdlib/now/4     1.05  1428.2±133.49ns        ? ?/sec    1.00  1363.3±65.90ns        ? ?/sec
stdlib/now/5     1.01  1391.0±117.71ns        ? ?/sec    1.00  1374.8±60.66ns        ? ?/sec
stdlib/now/6     1.02  1365.4±63.80ns        ? ?/sec     1.00  1344.3±46.68ns        ? ?/sec
stdlib/now/7     1.02  1387.7±76.32ns        ? ?/sec     1.00  1363.6±53.60ns        ? ?/sec
stdlib/now/8     1.01  1411.2±115.90ns        ? ?/sec    1.00  1398.2±92.98ns        ? ?/sec
stdlib/now/9     1.03  1416.0±93.12ns        ? ?/sec     1.00  1374.0±56.60ns        ? ?/sec

Closes #81

[loyd]

During development:

• The "test_reference_source_calibration" test fails (even on the main branch) with mean=1124.1603625000016
• The "test_recent" test fails sometimes (even on the main branch). Now, saturation hides the problem and the test passes always. Need to investigate, actually.
• #![deny(clippy::all)] is inconvenient during development, I suggest to add -D warnings in CI instead.

[tobz]

• The "test_reference_source_calibration" test fails (even on the main branch) with mean=1124.1603625000016

This test is sort of inherently flaky, so I wouldn't worry about it. I'll likely end up marking it as #[ignore] before cutting the next release.

• The "test_recent" test fails sometimes (even on the main branch). Now, saturation hides the problem and the test passes always. Need to investigate, actually.

Hmm, yeah, interesting. 🤔

[loyd]

Thanks! Waiting for the new release to move back to upstream instead of a fork.

[tobz]

FWIW, I tracked down and fixed the issue with the test_recent test, which was related to the multiple tests, test_recent included, interacting with the global recent time in a way that with the right ordering, it would change the value that was being asserted against.

[tobz]

@loyd This PR is now released as of quanta@v0.11.0.

Thanks again for your contribution. 👍🏻