Simplifies and standardizes compilation of the return stack words

[patricksurry]

• Introduces a new ST header bit to flag the six stack juggling words, removing the need to search strip-table for every word compilation. This speeds up compile, and saves a net of about 30 bytes.
• All six words use a standard 10 byte prologue in the non-native case to copy the jsr return address+1 to tmp1, so they can end with jmp (tmp1) instead of putting the return address back on the stack. This is slightly faster and means we don't need to top and tail different lengths for each word
• The 2r@ word can now be native compiled or not, just like the other five words.

[SamCoVT]

This looks a lot simpler/cleaner than the previous implementation.