Remove HIDE_MORPHO from compute_chosen_words

[linas]

compute_chosen_words() should be factored to remove the HIDE_MORPHO stuff, and the HIDE_MORPHO should be a distinct step.

The current implementation of HIDE_MORPHO in compute_chosen_words is not quite right: it assumes that only the suffix has connectors attaching to other parts of the sentence, and so fails to merge the disjuncts correctly if this rule is broken.

[ampli]

My current (i.e. in the Wordgraph version) implementation of HIDE_MORPHO removes only the LL links, and leaves the other links intact. However, I had to set the chosen_words correspondent to the hidden (Hebrew) morphemes to " " (blank) instead of NULL, else the linkage_print_diagram_ctxt() segfaults...

Words in chosen_words corresponding to stems are sill NULLed in my version, but if a stem would have other links then LL, the said segfault will occur. All of this can be fixed.

Since a general HIDE_MORPHO code is not trivial, it may indeed be a good idea to do it in a separate step.

[linas]

once the dict is built, it should not be modied -- if it is, that would cause multi-threading issues. Clearly no one is multi-threading .. but hey .. anway, many of these routines should be marked to have const arguments, to avoid such issues.