[aat] Add shaper-selection mechanism to prefer aat

[behdad]

Continuing discussion from #322

We want a programatic way to contro whether 'aat' shaping should be preferred over 'ot'. Basically, making the shaper Apple-style.

We want to use the existing shaper-selection mechanism. Ie. caller provides list of shaper strings to hb_shape_full(). Unlike current method, I don't want our shaper to fail if the requested method was not available. Ie. if user asks for "aat", we try aat and then ot and then whatever fallback. We would NOT fail if font doesn't have aat.

With that in mind. I like to propose the following changes:

• Keep "ot" to mean what it does right now: prefer OpenType and then try AAT and then fallback,
• Add "aat" to mean: prefer AAT, then try OpenType then fallback,
• Add "fallback" that doesn't try OpenType or AAT. This will make the current separate "fallback" shaper obsolete. Although, there's a benefit to current fallback shaper in that it's 200 lines of code, it's much easier to verify as secure and safe. But don't think we need that property.
• Add "auto" to mean whatever is default. Alternate namings could be "harfbuzz", or "default". I don't like "default" since it's too close to saying default shaper order (which is different concept).

[behdad]

Closing this as we found ways to be happy with whatever we have so far. Feel free to reopen if you disagree.