Cache default font

[ederag]

Caching default font is the last bit needed to accelerate savegif from 12s to 3.4s (fully fixes #194).
The timings now are

sim GifOutput:   3.543932 seconds (15.66 k allocations: 369.376 MiB, 2.02% gc time)
sim ArrayOutput:   0.486234 seconds (459 allocations: 2.530 MiB)
savegif:   3.139735 seconds (16.19 k allocations: 492.624 MiB, 0.02% gc time)

set_default_font(FTFont(path)) can be used to bypass any findfont call, if desired.
Please let me know if this is OK to export (I'd add a docstring then).

[rafaqz]

Thanks! We probably don't need to export, as DynamicGrids.set_default_font has much clearer meaning than set_default_font, and I guess it will be rarely used and then only once.

But you can add a docstring so its part of the unexported interface. Just include it using the full DynamicGrids.set_default_font string where it seems to fit in the docs.

[ederag]

Fully agreed, thanks !
It should be ready for further review.

[ederag]

On second thoughts, font-related error handlers needed an update
(now that FTFont and set_default_font can be used)
=>

julia> DynamicGrids._fontwrongtype(1)
ERROR: ArgumentError: font must be either a String or a FreeTypeAbstraction.FTFont,
got `1` which is a `Int64`.
...

julia> DynamicGrids._fontnotfounderror("cantarell")
ERROR: ArgumentError: Font "cantarell" wasn't found in this system.
Specify an existing font with the `font` keyword,
or call DynamicGrids.set_default_font first,
or use `text=nothing` to display no text.
...

julia> DynamicGrids._nodefaultfonterror(("cantarell", "sans-serif", "Bookman"))
ERROR: Your system does not contain any of the default fonts
("cantarell", "sans-serif", "Bookman").
Use DynamicGrids.set_default_font first.
...

The last commit is a proposition to move those error handlers to the end
(the beginning logic is then easier to grasp),
but I have no strong opinion about it.

[rafaqz]

LGTM. Sorry I took so long to review this!

[rafaqz]

Is there a reason these are wrapped so narrow now? 92 chars is the standard max width for Julia.

[ederag]

This was an application of "one chunk of thought per line",
which makes things much easier to read.
I agree that the newline before throw was unnecessary, a9dbfc8 is better ?