fix(DRM): renew the mediaKeySystemAccess on Edge and Firefox when using a Playready keySystem.

[Florent-Bouisset]

Edge and Firefox browsers tends to crash when reusing the mediaKeySystemAccess object when switching between different contents while using Playready DRMs.
I propose to recreate the mediaKeySystemAccess when switching content on those device to mitigate the issues.
This would have the disadvantage of not having a session cache for temporary licence, but It would improve the stability.

[peaBerberian]

Note to give some more context: it seems that the very new PlayReady support on Firefox is very buggy for now, yet it adds hardware DRM on firefox windows, so we highly do want it because it allows support of higher-quality content with Firefox windows (like only edge windows does for now).

What we noticed is that:

1. Changing between two different contents (encrypted with different keys) with PlayReady SL3000 led to a GENERATE_KEY_REQUEST_ERROR (an error while asking to generate the challenge) when switching to the second content

2. Reloading the same content (or different ones encrypted with the same key I guess?) with PlayReady SL3000, with our active MediaKeySession cache (so no need to reload the licence) led to infinite rebuffering

3. Our usual work-around for those kind of lower-level CDM problems, which is to restart from the MediaKeys-creation step, weirdly didn't have an effect here

4. However, what is proposing Florent here: re-starting from the first EME API, navigator.requestMediaKeySystemAccess, (which until now was always assumed to mainly be about polling decryption capabilities, so no need to do it X times) did fix those two issues

So we talked about it and decided that for Firefox PlayReady, we can just restart from the beginning and not profit from our MediaKeySession cache - nor on the optimization of not re-calling EME API (which can last hundreds of ms even on performant devices) for already-loaded contents.

---

In an almost-related issue, we also had a lot of UWP (universal windows platform - which relies on Edge for Web views) and Edge windows issues recently (well, since forever to be perfectly honest).
We decided in the same effort to also disable the cache - from that same step - for Edge PlayReady as we figured the optimizations it allows does not justify all the issues it brings.

[github-actions]

✅ Automated performance checks have passed on commit 0ce0dcfc5cd7866eda7069f0d21c3d86ad019914 with the base branch dev.

Details

Performance tests 1st run output

No significative change in performance for tests:

Name	Mean	Median
loading	19.67ms -> 19.66ms (0.007ms, z: 0.14609)	27.00ms -> 27.00ms
seeking	18.13ms -> 14.07ms (4.061ms, z: 1.54321)	11.40ms -> 11.40ms
audio-track-reload	27.12ms -> 27.00ms (0.124ms, z: 0.55687)	39.30ms -> 39.45ms
cold loading multithread	49.02ms -> 48.11ms (0.913ms, z: 9.82316)	71.70ms -> 70.65ms
seeking multithread	22.74ms -> 26.76ms (-4.021ms, z: 0.40080)	10.50ms -> 10.50ms
audio-track-reload multithread	26.92ms -> 26.89ms (0.031ms, z: 0.14580)	39.45ms -> 39.45ms
hot loading multithread	16.07ms -> 15.91ms (0.157ms, z: 3.17029)	23.10ms -> 22.95ms