Note: For now, the scale cannot be chosen. The input is always quantized
to the nearest 12-tone equal temperament step.

The version of the description added in the previous patch was
marked up wrong, resulting in the text appearing in two separate
boxes rather than one box with two paragraphs inside it.

This patch fixes that issue.

Note: For testing purposes, these sysexes currently affect the live config
directly. A later commit will change them to affect the config in flash.

As the config struct is about 2KB in size, and we end up needing ~2 copies
on the stack in cmd_0x1A_read_quantizer_config_(), we need to increase the
stack size to 8KB to prevent overflow. This can likely be reduced in future.

The previous commits needed more stack size than was previously allocated,
and changed the STACK_SIZE variable in configure.py accordingly.
Strangely, however, the linker still only allocated 2KB to the stack.
On further investigation, the line

    STACK_SIZE = DEFINED(__stack_size__) ? __stack_size__ : 0x800;

doesn't work, it always sets STACK_SIZE to 0x800. But changing it to

    STACK_SIZE = __stack_size__

works perfectly

It is important to ensure that the addresses and lengths of the existing
NVM sections (settings and lut) are not changed by this patch.
I have manually verified this as follows:

* Build the firmware before applying this commit
* Run:
    $ objdump -t build/gemini-firmware.elf | grep _nvm
* Then apply the commit and rebuild
* Run objdump again
* Compare the two sets of outputs. The relevant columns are the first
  (symbol value) and last (symbol name)