Coerce debug config to boolean

I toggled "true" to "false" and it kept debugging. I should've just
removed the line, I guess, but I didn't want to.

Make it possible to actually set concurrency

We were reading it from the top-level of config.json, but it's being
filtered out from the yaml and not being included, so we always have the
default of 2.

I'd like to try going to 1 to help investigate something else, and now
it will be possible.

Also, now that we're able to set this value, we need to make sure it's
not a string.

Update example yaml to use colon hash syntax

Consistent with [Help Doc](https://docs.codeclimate.com/v1.0/docs/duplication#section-enable-the-engine).

Raise explicit config errors

When presented with invalid config we can't establish any semantic
meaning is, it is better to fail noisily and inform the user than to
proceed silently & probably not do what the user intended.

add count_threshold option (codeclimate#142)

* add count_threshold option

adds the option to only check duplication if the code is found in more than just two locations.

* make count_threshold_for less confusing

* allow changing global default count threshold

* make count threshold default no longer language-specific

* spec for count_threshold_for

Disable debug output by default

This was helpful for identifying potentially problematic files, and
helped lead us to identify minified js files as a problem, but it hasn't
been as helpful recently, and can actually lead to exceeding the maximum
output limit for repos with many files!

Update to latest jruby

This is motivated by a rare error which was fixed in 9.1.3. Some more
context in this issue:

jruby/jruby#1279

Double the stack memory limit

We're seeing some reports that analysis is failing with output like
this:

```
Error: Your application used more stack memory than the safety cap of 2048K.
Specify -J-Xss####k to increase it (#### = cap size in KB).
Specify -w for full StackOverflowError stack trace
```

Refactor away from OpenStruct for php AST

I believe this is the root cause for many of the OOM errors we've been
seeing.

Our original analysis pointed toward:

1. repos with big minified js files (resolved by skipping those)
1. php-only repos (mysterious...)

Since ripping off the bandaid Monday and surfacing OOM errors to
end-users, we've been hearing about several PHP-only repos that are
erroring, so I took some time to read through the php parsing code
looking for a possible memory leak.

Eventually I noticed that we're using OpenStruct to represent our "Node"
concept for the php AST that we build. I've heard that OpenStruct isn't
very memory-efficient or fast and decided to try refactoring away from
it.

I'm able to reproduce the OOM error via the CLI on my clone of
etsy/phan, and the error no longer occurs with this refactor.

:tada: :tada: :tada:

Guard against divide by zero error (codeclimate#145)

* Guard against divide by zero error

Fixes a bug where empty js file could cause an analysis to error