Add mechanism for catching exceptions that has limited use even with -fno-exceptions, as well as a mechanism to detect when a destructor is called during unwind and to avoid throwing in these situations.

totalSizeInWords() method on generated struct readers/builders returns total size of the struct, not counting far pointer overhead.  Useful for allocating space for a flat copy.  Also a couple misc tweaks.

Implement ability to set() struct and list fields by deep-copying from a corresponding Reader.  This copy requires no schema knowledge.

When an under-sized struct is found while traversing a builder, upgrade it to the expected size.  This is needed to allow easy copying from readers to builders.  Next, this will need to be implemented for lists as well.

Restore ability to upgrade from List(Bool) to List(T) where T is a struct whose @0 field is of type Bool.  I previously disallowed this to reduce complexity, but it turned out to actually increase complexity.

Delete Code Friday:  Inline structs/lists/data are too complicated for not enough benefit.  There goes a week of my life that I'll never get back.

Implement 'Object' field type, which can be any pointer type.  The caller specifies the type when calling the accessors.

Correctly handle inline fields that aren't present in the parent struct (i.e. because we're reading an older version of the struct that lacked those fields).

Improve unions.  Members of a union are now scoped within the union, although they share ordinal space with the parent struct.

Replace most CAPNPROTO_ASSERTs with the new logging/error macros.  Also add a SYSCALL macro and use it.  Also add commentary.

Update implemented pointer format to match docs:  Don't store field count in struct pointers because it isn't needed.  Instead extend data and pointer section sizes to 16 bits, which should be enough for anyone.  Also, store offsets from the end of the pointer rather than the beginning, because this makes it possible for them to be zero, which compresses better.  In particular, the root reference will always have a zero offset, so this will shave a byte off of very small packed messages.