You may be underestimating the amount of work involved.
Translating the raws and other external files will get you perhaps 60% of the way there, at least for languages using the Roman alphabet. Unfortunately, doing so will involve creating a system to switch the raw directory based on a selected language. Worse, certain key bindings should also be translated, but would have to be loaded before the user could select a language, or re-loaded on language selection, unless the language could only be selected through editing the init.txt file.
Translating the hard-coded strings will contribute perhaps another 25%, but will require something like
gettext for on-the-fly translation.
Another 5% will involve converting various routines to deal with quirks in each language. Things like inserting articles into names and descriptions currently make certain assumptions that don't hold true outside of English. A new tag might be required, specifying the grammatical gender of every object in the game. This can be eased slightly by moving more of the hard-coded items, buildings, and materials into the raws, but that alone is a messy project.
The last, hardest 10% will be changing the output routines to use true Unicode and/or a large enough tileset to support non-alphabetic languages, and making every potentially conflicting code point configurable. Again, rawification will help here.
Worse, the translation effort itself will involve someone fluent (or at least proficient) in each target language. Given the way releases are handled, these language packs will probably be distributed externally, lagging some time behind the official release. Meanwhile, the bug tracker will be inundated with complaints that this or that language doesn't work right, or that some generated sentence is ungrammatical in certain situations. This is exactly the kind of stressful situation Toady has decided to avoid by not catering to third-party interfaces.
For a quick gauge of the effort required, try to combine the English, Chinese, German, Japanese, and Spanish versions of NetHack into a single executable. Then translate it into Arabic.
That said, it would be nice. When the source goes public, I have no doubt that someone will try it. Meanwhile, this becomes another argument for moving hard-coded stuff into the raws when possible and convenient.