Improve tilecache performance

[jcupitt]

This test prog:

#!/usr/bin/env python3

import sys
import pyvips

image = pyvips.Image.new_from_file(sys.argv[1])

image = image.tilecache(tile_width=image.width,
                        tile_height=256,
                        max_tiles=1,
                        persistent=True)

patches = [image.crop(x, y, min(256, image.width - x), min(256, image.height - y))                    
           for y in range(0, image.height, 256)
           for x in range(0, image.width, 256)]
           
for patch in patches:
    pixel_array = patch.numpy()

With this (7 gb) test tiff:

$ vips bandjoin "st-francis.jpg st-francis.jpg st-francis.jpg" x.tif[tile]

I see:

$ VIPS_CONCURRENCY=1 /usr/bin/time -f %M:%e ./readtile.py x.tif
472656:13.37

If I change this line:

image2 = image.tilecache(tile_width=image.width,

ie. create the tilecache but never use it, I see:

$ VIPS_CONCURRENCY=1 /usr/bin/time -f %M:%e ./readtile.py x.tif
143064:4.51

It looks like the tilecache alone is taking almost 10s of runtime. Can we improve this?