Fix tick label rotation and layout issues

[nagix]

There are a few problems ticks.minor and ticks.major configuration:

1. Can't update scale.ticks.* options at runtime because the scale class copies the minor and major ticks options from ticks.* on the initial chart creation, and it doesn’t update them on the following update calls as the values already exist ([BUG] Can't update scale.ticks.* options at runtime since 2.7.0 #4896).
2. ticks.major.enabled is set to false by default, but ticks.major options are effective. The behavior should be consistent with the ticks.major.enabled setting.
3. Only ticks.minor options are used for calculateTickRotation, fit, draw and _autoSkip calculations, so major tick labels can overlap or be cropped.
4. Tick labels on X axis overlap each other. This happens because calculateTickRotation only considers the first tick interval before fitting, and doesn’t take into account margins for the tick labels on the left and right.

This PR fixes these problems as follows.

1. Deprecates mergeTicksOptions and adds parseTickFontOptions to parse minor and major tick font options. Omitted options are inherited from ticks at runtime.
2. Uses minor tick options when ticks.major.enabled is false.
3. Deprecates computeTextSize and adds computeLabelSizes to measure tick labels and find the widest, highest, first and last labels at a time.
4. Improve calculateTickRotation so that it takes into account margins for the tick labels on the left and right.

Edit: This PR fixes the following problems as well.

5. The proposed calculateTickRotation calculates the maximum scale height using the scale label size, the tick mark size and the tick padding to remove the overlap with scale label ([BUG] Tick label overlaps axis label, long labels cropped #5906).
6. calculateTickRotation has a loop to determine the label rotation and it can be repeated 90 times at most. The proposed version eliminate the loop to improve performance.

Edit2: Problem 1 and 2 are addressed in #6102.

Master: https://jsfiddle.net/nagix/amtLsdfb/

This PR: https://jsfiddle.net/nagix/mas98q17/

Fixes #5906
Fixes #2106

[kurkle]

Would it be better to clip the labels instead of taking available height into account? At least when autoSkip=false? (Not something to do in this PR)

[etimberg]

I think I understand how this works. I haven't had a chance to test it out locally yet.

[kurkle]

Small performance increase in this PR, rendering 1M points on timescale.
These are single runs, so not absolute. But the difference in total time seems to stay above 1000ms repeatedly.

Master:

{
  "1 init": 1.8,
  "2 update/ds/0": 782.6,
  "2 update/ds/1": 1079.2,
  "2 update/ds/2": 748.4,
  "2 update/ds/3": 727.7,
  "2 update/ds/total": 3338.3,
  "2 update/layout": 4393.1,
  "2 update/total": 8190.5,
  "3 render/draw/ds/0": 105.6,
  "3 render/draw/ds/1": 111.3,
  "3 render/draw/ds/2": 119,
  "3 render/draw/ds/3": 119.9,
  "3 render/draw/ds/total": 456.2,
  "3 render/draw/total": 458.1,
  "3 render/total": 476.5,
  "x from page load": 8673.9
}

This PR:

{
  "1 init": 1.7,
  "2 update/ds/0": 822.4,
  "2 update/ds/1": 824,
  "2 update/ds/2": 1046.8,
  "2 update/ds/3": 794.2,
  "2 update/ds/total": 3488,
  "2 update/layout": 2910.7,
  "2 update/total": 6803.2,
  "3 render/draw/ds/0": 107,
  "3 render/draw/ds/1": 108.6,
  "3 render/draw/ds/2": 118,
  "3 render/draw/ds/3": 123.2,
  "3 render/draw/ds/total": 457,
  "3 render/draw/total": 459.1,
  "3 render/total": 477.9,
  "x from page load": 7288.3
}

250k ponts per dataset. something breaks in chrome (render process dies) when putting 300k points in a set.

[nagix]

@kurkle Thanks for the analysis. It seems that the performance gain comes from "2 update/layout", so this is what I expected. Does the master even crash in Chrome?

[kurkle]

@nagix both crash, so not an issue in this PR. I think maybe path becomes too long.