Fixes #1.

Update the Makefile to automatically fetch theos (git submodules!) if it's missing. Update the README to contain compilation instructions and remove the erroneous mention of a local Theos.

Convert libpng and zlib to statically-linked subprojects of pincrush (instead of aggregated individual subprojects that are manually linked in.)

Only use the included libz if we're building for Windows. Also, wow, this is an insanely ugly hack. Remind myself to look into why adding to xxx_OBJ_FILES does not work but modifying everything else does.

Add linux and windows support and a bunch of theos modifications that will never be fed back to trunk in this form.

Apple's CgBI format is insane. Why is it 0x02 for anything-with-alpha and 0x06 for everything else? Fixed.

Add multiple levels of verbosity: info, ridiculous, and insane. Display some memory usage statistics in insane mode.

Argh. I forgot GREY_ALPHA, so images of that type were not getting upconverted, which caused rowbytes to be wrong (well, reading in 2bpp and writing out 2bpp-as-4bpp is a bad idea :P) which caused crashing. Thanks to Jay.

Allocate extra slack space at the end of each row (16 bytes' worth) to alleviate bizarre chunked mode crashing issues, and flip the interlace pass loop inside-out to match pngtest.

Instead of reading the whole image in and processing it, read and write chunks of rows (default: 8 rows, configurable) to massively save on memory for larger images (a 320x480 image takes ~614kb of memory, with an 8-row window we only use 10kb at a time.)

Hack Theos's library target so it doesn't generate a dylib at all. This really should be its own target.

Move the crushing logic into its own crush() function (which takes in/out filenames), and fix the various memory and file pointer management maladies I was committing. Pincrush doesn't leak memory (according to valgrind), even for multiple files.
This allows us to process multiple files with a single invocation, such as pincrush -i file1 file2 file3.