Improve support for empty and non-ASCII-alphabetic properties, enum values, and definitions

[ucarion]

This PR fixes #16 as well as related issues.

At its base, this PR does two things:

1. For TypeScript properties in particular, where the name assignment system isn't used at all, this PR adds code that detects properties that are potentially invalid TypeScript identifiers, and escapes those property names.
2. For all other targets and cases (including TypeScript), this PR alters the identifier assignment code to better handle generating names from strings like "", "1foo", " ", or "\uFDFD".

The name generating logic now has the following invariants:

1. Non-ASCII alphanumerics are always considered "whitespace", in the sense that they are stripped from names and are considered the boundary of a logical "word".
2. Logical "words" can only start with ASCII alphabetics. Leading ASCII digits in words are always stripped.
3. If a name has no non-empty logical "words" within it, then the name will be generated from the logical words "default" and "name", producing default_name, defaultName, etc. depending on the target's chosen casing.

Test cases that exercise these cases are added to all targets. However, Java+Jackson does not support empty property names or enum values, and Go does not support non-ASCII-alphanumeric property names. For this reason, each target's test cases are handled separately. All targets are required to support non-ASCII-alphanumeric definition names, however.