Fix more panics following PR #51

[toomim]

From #51:

I've fixed all the panics that this test found. I suspect there are more lurking around - might be good to strengthen this test.

Here are fixes to some more that were lurking. Strengthened the test too. It now
tests byte replacement, deletion and truncation as well as insertion, over three
shapes of document instead of one.

One of the fixes is to the #51 fix itself. The n == 0 guard on local parents
rejects a zero offset, but next_time - n still underflows when n is larger
than next_time.

The wasm part is separated into its own commit because it changes a visible JS API: fromBytes used to unwrap, creating an illegible fault. Now it throws.