[3.x] Add readahead to VariantParser

[lawnjelly]

Adds a readahead buffer to VariantParser, to prevent large numbers of freads for single bytes, which is inefficient.

Notes

• Part of a few areas I'm aiming to improve as part of profiling during the MergeNode PR. VariantParser appears in a lot of places in Godot, and it turns out that it ultimately calls fread byte by byte (at least on linux), and this can be super slow, even with disk caching.
• The VariantParser always seems to read in linear order, so isn't subject to some of the pitfalls of readahead (moving the filepointer around etc). This type of access also can be optimized in a way that freads cannot (the file could be changed between each individual fread, for instance).
• There are alternative approaches to the same problem, such as putting the readahead into the file classes, but this could be counterproductive in some situations without knowing the access patterns. Another alternative is reading the whole file into RAM first, but this could use up a large amount of RAM without necessarily being significantly faster than a smaller readahead.
• This gives about 30% speed increase to loading scenes (it varies according to the scene content), especially useful for large scenes. This is measured using scenes from WroughtFlesh and TPS demo, and SSD. Increase in performance may be higher using hard disk.

Measurements

For performance measurements I used a version where the new and old could be #ifdefed between the two, and added a timer around the load_scene() in EditorNode::open_request().

[timothyqiu]

Looks good to me :) Left a few small nitpicks.

[timothyqiu]

I'm a bit surprised by the recursive call TBH 😛

Could be replaced by the following structure. But it's also fine to keep it the way it is.

if pointer >= filled:
    filled = _read_buffer()
    if filled == 0:
        pointer = 1
        eof = true
        return 0
    pointer = 0

return buffer[pointer++]

[lawnjelly]

Yeah, same result.

One of the reasons I got into the habit of putting the expected result to be the if statement (rather than the else) is that long ago I read the assembly generated tends to favour that branch (much like the unlikely keyword, before that was available). Whether that is still the case I don't know, have to test in godbolt! 😀

[akien-mga]

reduz approved the 4.0 PR, so this one should be good too.

[akien-mga]

Thanks!