The output of `getBindings()` is largely just a list of the class names
registered in the container, but it doesn't contain information on how
those classes are actually configured. As such, changing `ecs.php` in a way
that updates sniff / fixer rules would not invalidate the cache, meaning
those new settings would not be checked against any unchanged files.

My approach here still isn't perfectly reliable, but I think it's a usable
middle ground. We still report all the registered classes, but
now I explicitly pull out the configuration data for sniffers and fixers
to add to the serialized data. Then I also hash out the settings used for our
sniffers, to support the upcoming PHPCS rulesets / standards support.

The ECSConfig previously constructed as-needed, is now provided via DI.
This is required for the upcoming feature, but has the added benefits of making
sure the cache hash uses the hash for the exact same config that will be used
for sniffing *and* improving the performance of sniffing
(since every file is currently creating a config, triggering filesystem ops).

Closes easy-coding-standard#30

First, the class has been relocated to a more
idiomatic directory location, matching it's
namespace. Then many new tests were added to cover
different types of cache invalidation that can
occur. I also added a positive control test to
make sure caches *aren't* invalidated when we
don't want them to be.

You will want to read the associated class comment
for information on why it's structured so
strangely. The existing cache test was removed,
since it didn't work correctly Again, see the
class comment for why.

In the course of testing, I discovered the previous approach didn't work
for third-party sniffs.

The two needed changes:

1. "Real" fixers following the PhpCsFixer intended implementation with
`AbstractFixer` have the `$configuration` property pulled explicitly instead
with reflection-based visiblity overwriting. The serialized leading `\0\0`
approach seemed too non-obvious for future maintainers on a second viewing.
2. The condition was wrong. I don't know what I was thinking at the time. :|

…tandard into kesills-cache-invalidation

… hashing

Left a comment detailing not to use `::$configuration` like I had attempted to,
so future maintainers don't make the same mistake.