Change Taiko pp to be more consistent with stable

[cory2067]

Fixes #4275

Currently, if you run osu-tools to calculate pp on a taiko score (especially one that's not an FC), the result would be wildly different than what we observe live.

This is due to these 2 factors addressed in this PR:

• ppy/osu has combo scaling. This is inconsistent with live pp, which doesn't care about your combo at all
• OD calculation off by 0.5ms (which made all scores a few pp too low). ppy/osu is using 50ms window for a 300 on OD0, but stable (for whatever reason) uses 49.5ms

However, some scores are still off by a couple pp. I think it might involve the star rating being slightly different on certain maps. (example which is 4.20* on stable/web but 4.26* on lazer). This PR doesn't touch the star rating/strain at all.

Here's the output of osu-tools running profile Cychloryn -r 1:
On master (off by 185.7pp): https://pastebin.com/JWy3ZBHP
Proposed (off by 1.3pp): https://pastebin.com/haQ4bXdS

[smoogipoo]

Re: combo scaling, see: https://github.com/ppy/osu-performance/blob/585e2079832253f2ecf3166cb1e74a7daf4879a6/src/performance/taiko/TaikoScore.cpp#L73 this is currently live

[cory2067]

Ok, there's something strange going on here -- either that code has a bug or is not actually live.
There are plenty of counterexamples that point to combo scaling not being a factor:

https://osu.ppy.sh/beatmapsets/353583#taiko/1594500

The second score missed basically dead center in the middle of the map.
But these similar scores have almost the same pp.

If combo scaling was actually working as the code seems to intend, then vzk's pp should have been multiplied by sqrt(1169) / sqrt(2320) = 0.71

[cory2067]

bump on this, as combo scaling does not appear to be live so we should fix the inconsistency

maybe it is live, but there could be a bug that causes the max combo here to be <= 0 and therefore not do anything (just speculating)

[chayleaf]

The "problem" could be the fact combo in taiko pp system score's combo is accounted for MaxCombo - MissCount, not sure if that's the case but I'm pretty sure it is.

[chayleaf]

Again, new version https://docs.google.com/spreadsheets/d/1TuKl2XP7Pz66YUhXrO1sXs9zdtCDWkpxcKjA8ScIjQA/edit?usp=sharing
An issue is repeating maps being underweighted even if they aren't as repeating as converts.

[smoogipoo]

Going to close this one for now until ppy/osu-performance#117 is merged, after which a similar change will be made here (minus the GreatHitWindow change).