Fix the grid, the bottom strip and scrolling when the font is large

[demoj1]

Thanks for the package, it is lovely to use.

I hit three bugs on a HiDPI screen with a large font (Iosevka, 19.6pt, so one
character is 13px wide). They all come from the same thing, so they are in one
PR.

The root cause

The package uses two kinds of pixels:

• canvas pixels — the numbers inside the SVG;
• window pixels — what Emacs draws and scrolls.

When nothing scales the images, both are the same number. So the mix-up never
showed. As soon as a scale factor appears, three things break at once.

Bug 1: the grid does not line up with the day names

The header image is made with :scale 1. The grid tiles are made with
create-image and no :scale. So Emacs scales the tiles with
image-scaling-factor. Its default is auto, which means char-width / 10.
On my setup that is 1.3.

So the grid is 30% wider than the header. Both use the same column-width, so
every column moves a bit further away from its day name. By Sunday the gap is
about a third of a column.

You can see it in the numbers: the day names are 187px apart, the grid columns
are 245px apart, and 245 / 187 = 1.3.

Fix: a new option org-timegrid-image-scale (default 1.0). Every image
now gets the same explicit :scale. The SVG is laid out in
window-width / scale, so the canvas still ends exactly at the window edge.
Mouse clicks and the fringe come in window pixels, so I convert them before
hit-testing.

The option is also a zoom control. 1.3 gives the old, bigger grid — but now
the header and the events grow with it.

Bug 2: a white strip under the last hour

The canvas is 1302px tall: 24 hours × 60 × 0.9 px per minute, plus a 6px inset
at the top. One tile is one hour, so 54px.

ceiling(1302 / 54) is 25. So the last tile is only 6px tall. But a 6px
image still takes a whole text line, so it looks like a white strip across the
calendar. The newline after the last tile added an empty line under it.

Fix: round the tile count down and give the extra 6px to the last tile, so
it is 60px instead of 54 + 6. A new org-timegrid--tile-bounds returns the
bounds, and the image map uses it too, so clicks still work on the whole
canvas. The last newline is gone.

Bug 3: scrolling and org-timegrid-recenter stop at the wrong place

window-vscroll and window-body-height return window pixels. But the cursor
position, org-timegrid--image-height and the tile offsets in
org-timegrid--window-scroll-pixels are canvas pixels.

At scale 1 both are equal, so this was fine. At 1.3 every scroll lands 30% off,
org-timegrid-recenter puts the cursor far from the middle, and the maximum
scroll stops before the end of the day.

Fix: convert at the border. Three small helpers do it: --display-y,
--display-tile-height and --display-height. The last one adds up the tiles
instead of scaling the canvas once, because each tile image is rounded to whole
pixels on its own and the error adds up (1688px, not 1693px, over 24 tiles).

Extra: org-timegrid-scroll-style

This one is not a bug fix, just a small option. Today the view scrolls as
little as possible, so after moving for a while the cursor sits at the edge of
the window. org-timegrid-recenter fixes it by hand.

The new option lets you choose:

• minimal — today's behaviour, and the default;
• center — keep the cursor in the middle while there is room to scroll.

The centring code now lives in org-timegrid--center-cursor, and
org-timegrid-recenter uses it too, so the maths is written once.

Happy to drop this last commit if you want the PR to be fixes only.

Notes

• Nothing changes for users whose character is under 10px wide, because
  image-scaling-factor already returned 1 for them. Both new options default
  to the current behaviour.
• The commits build on each other, so it is easiest to take them in order.
• ERT: 20 tests pass. Two are new: one for the tile bounds, one for the scale
  option with bad values.

[Gleek]

Hey! thanks for the PR.
I was actually already working on some font scaling changes (2a03d9d) and they cover few of the bugs you reported already.

Can you please see if there are still issues with alignment after this commit

I think the other fixes might still be valid. Can you please create separate PRs for individual features and bug fixes so that reviewing is easier on my end.

[demoj1]

Thanks for the quick reply, and for the zoom work.

I checked 2a03d9d. The alignment problem is still there. Measured with emacs -Q, an empty backend and only the font set (Iosevka 19.6pt, so a 13px character):

	step
day names in the header	136px
column lines in the grid	176px

176 / 136 = 1.29.

The reason is that your zoom and image-scaling-factor do different things and multiply. Your zoom makes the drawing inside the SVG bigger. image-scaling-factor stretches the finished image afterwards, and it only touches the tiles, because the header already passes :scale 1 and the tiles do not. Adding :scale 1 to the tiles brings the columns to 135px, in line with the header.

The other two problems are also still there:

• org-timegrid--tile-count uses ceiling, so with a 1302px canvas and 54px tiles the last tile is 6px tall. A 6px image still takes a whole text line, so it draws a white band under the last hour.
• Scrolling mixes units, but only because of the scaling above. Once the tiles are not stretched, canvas pixels and window pixels are the same number again, and it fixes itself. So I dropped that patch.

I split the rest into separate PRs as you asked:

• Do not let Emacs rescale the grid tiles #2 — :scale 1 for the tiles (the alignment fix)
• Give the leftover canvas to the last tile #3 — give the leftover canvas to the last tile (the white band)
• Follow Emacs' own scrolling variables when moving the cursor #4 — org-timegrid-scroll-style, an option, not a fix

Each one is on its own branch off current main and stands alone. Closing this PR in favour of those.