remove timestep from Q

[florianmattana]

Applied the fix removing ts from the Q scale computation. Here are the results:

Fix	Mean diff	Max diff
Baseline	0.168081	1.693359
Symmetric range	0.168939	1.614502
Remove ts from Q scale	0.166455	1.372681

The mean difference barely moved, but the max dropped from 1.69 to 1.37. It seems ts was causing spikes on specific values rather than a uniform error across the board. That makes sense since ts only distorts the scale factor, and the impact depends on how far each value is from the tile's max.

Let me know if we merge this first or should instead work on the data itslef

[ParagEkbote]

Thanks for taking the time to diagnose the timestep scaling issue. Feel free to test with data if needed to get more verifiable results.

I had a query. Should the approach in #7 work alongside this PR to reduce the high mean error, instead of being a seperate approach?

cc: @florianmattana