Avoid measurement dead-spot between sync and async phase

[camillobruni]

We could potentially miss a gc between the performance.mark calls for syncEndTime and asyncStartTime.

This PR changes the code to reuse the sync end times and marker for the async start.
This means we'll now also measure one additional performance.mark call.

An alternative would be to replace

• const syncEndTime = performance.now();with
• const syncEndTime = performance.mark(syncEndLabel).startTime;
  to have the same timestamp and one call less.

Summary 5 runs:

browser            Google Chrome          Firefox                Safari
version            131.0.6778.140         133.0                  18.1.1 (20619.2.8.11.12)
Score Before       32.07 ± 0.67%          27.228 ± 0.13%         32.75 ± 0.48%
Score  After       31.14 ± 0.54%          26.10 ± 0.79%          30.36 ± 0.68%

I have to re-run the detailed measurements tonight when my machine is not in use to reduce noise.

[rniwa]

Seems sensible to me.

[julienw]

This looks reasonable to me as well.

[camillobruni]

Just rerun the numbers and there is no measurable change.

browser       Google Chrome          Firefox                Safari
version       131.0.6778.205         133.0.3                18.1.1 (20619.2.8.11.12)
Score Main    32.60 ± 0.58%          28.66 ± 0.39%          33.59 ± 0.59%
Score PR      32.54 ± 0.52%          28.63 ± 0.61%          33.51 ± 0.99%