Faster package downloads via S3

Checksums are optional and specified as anchors on package URLs.

Minor README update

Switch 2.0.0-dev and 1.9.3-dev to git since the svn->git mirror is back up.

This reverts commit 75f98270.

update README examples in order to use MRI last stable version

fix 1.9.3-p327 definition

added 1.9.3-p327

Rationale:

Both in development and in production, some usage patterns of ruby-build
are slowed down by the download phase. In scenarios such as
troubleshooting failing builds or with provisioning situations (chef,
vagrant...) the repeated download is unnerving, bandwidth wasting and
simply against etiquette towards tarball hosters.

It also happens that some source sites happen to be down and in such
cases it is helpful to be able to sideload sources to rbenv.

Behavior:

By default nothing changes.

If the variable CACHE_PATH is set, then ruby-build will use that
directory to store a successful download, and will check before
downloading if the tarball is already there, in which case downloading
is skipped.

The file is first downloaded as before in the tmp subdirectory and only
moved afterwards, thus ensuring consistency.

There is no default cache path and the optional variable is to be set by
hand, ensuring people know what they're doing when using ruby-build.

Additionnally, rbenv-install will helpfully set CACHE_PATH if and only
if a RBENV_ROOT/cache directory exists. Again, the directory has to be
created manually.

The CACHE_PATH variable internally ends with a slash to mutualize
non-cached cases. Still, consistency is ensured whether or not a slash
is provided externally.

Notes:

I'm not quite sure CACHE_PATH is a good name, maybe
RUBY_BUILD_CACHE_PATH is better and less conflicting.