Parse <sample> with base64 data

[paulfd]

Also added the necessary plumbing to load from RAM.
It's missing AIFF, and metadata reading since those API are heavily file-based for now.
Note that this is not exactly what's proposed in #91, since the goal is also to allow to "copy" these embedded samples as printable characters (e.g. in an edit box on the web, or a pastebin).

The syntax is as follows:

<region> sample=test.wav
<sample> name=test.wav encoding=base64 data=[... Base64 data ...]

Samples like this will be loaded fully in RAM; I made some modifications to the file pool to check both "fully loaded" and preloaded files on requests, which wasn't the case before.
The encoding=base64 is required to try and keep a compatibility path with the Rapture encoding mentioned in #91 down the line.
When reading an opcode value, the parser now behaves differently if the opcode name is data: it will ignore comments since base64 has the / sign, and will only stop on EOF, a new < character, or a new # character.
The underlying base64 decoder ignores whitespace, and can support optional padding with =.

Now that I think about it and write it down, it's not going to be compatible as is with #91 because the parser for the latter stops at a different character ($). Maybe a different opcode name (e.g. base64data= and data=) to allow the parser to parse these items differently? @jpcima, your thoughts?

Missing or would-be-nice, possibly for further PRs:

• Changes depending on the above discussion.
• Rewrite the metadata reader to allow to read from memory instead of just files. The way to proceed is probably to have a new provision in the abstract AudioReader to read the raw data directly, regardless of the source. This part was not obvious when I refactored this so for the time being only basic metadata will be read mostly for WAV files. Once done, the path parameter added to the AudioReader constructors is unneeded.
• Expand the AIFF library with a memory version.
• Read gzipped/bzipped data, which would reduce a bit the size of the base64 fields.

[paulfd]

I changed it to read the base64data field in the end, and updated the metadata reader, as well as adding a bunch of tests. Should be all good.