output: improve anisotropic filter rendering

[rr-]

Checklist

• I have read the coding conventions
• I have added a changelog entry about what my pull request accomplishes, or it is an internal change

Description

Resolves #1507. Improves #902.

For a definite solution, we'd need to artifically add a border around each texture face inside texture atlases, hence I keep #902 as open for now.

[rr-]

Actually, hold on – this makes the issue worse in the Gym near the staircase to the piano room. The problem is that the walls are constructed like this:

##
####
######
###### | ######
###### | ######
###### | ######
###### +-------
###### | ######
###### | ######

Since there's 1 quad for the left side and 2 quads for the right side, this causes the calculations to misalign the faces along the | seam.

[lahm86]

Actually, hold on – this makes the issue worse in the Gym near the staircase to the piano room. The problem is that the walls are constructed like this:

##
####
######
###### | ######
###### | ######
###### | ######
###### +-------
###### | ######
###### | ######

Since there's 1 quad for the left side and 2 quads for the right side, this causes the calculations to misalign the faces along the | seam.

I was just testing a TRLE and it's a similar problem there too, and with a skybox present the gaps are much more obvious. I'd imagine this kind of texturing is quite common.

[rr-]

Restored high precision room vertex computing – I don't think this has downgraded the original fix.