Refactor workaround for Okapi in-memory document handling

[wadimw]

Split off from #1049

This PR removes the URI injection using reflection on RawDocument which is potentially fragile. Instead, it replaces it with an override on QualityCheckStep START_DOCUMENT event handler that injects the expected value directly into the StartDocument resource through its public method StartDocument#setName.

Additionally, it slightly refactors the empty RawDocument approach to make it more explicit that the content of this document is not relevant - it's only used to "drive" the Okapi pipeline and provide Locale configuration.

---

Rationale for the URI refactor:

Force setting URI on a RawDocument backed by InputStream or a CharSequence breaks its contract as specified here:

The RawDocument object has one (and only one) of three input objects: a CharSequence, a URI, or an InputStream.

Even the private runtime of RawDocument itself depends on whether the URI field is set (e.g. here), meaning this workaround might introduce unexpected behaviour and is potentially fragile.

After removal of the URI injection from Mojito code, the only thing that breaks seems to be QualityCheckStep - specifically, it throws a NullPointerException on the following line:

@Override
public void processStartDocument (StartDocument sd,
	List<String> sigList)
{
	currentDocId = (new File(sd.getName())).toURI();
// ...

This can be avoided by setting an override on our existing QualityCheckStep subclass which will set this name if it's null. This way, the workaround is localized only to the class that actually needs it, rather than affecting all steps of Okapi pipelines that would rely on RawDocument.

[wadimw]

TODO should probably also get rid of

mojito/common/src/main/java/com/box/l10n/mojito/okapi/steps/FilterEventsToInMemoryRawDocumentStep.java

Lines 31 to 32 in f7a5979

	private File outputFile;
	private URI outputURI;

in this PR