added a really fully typed translation as a fallback for Metis, in rare cases where Metis correctly proves a theorem but has type-unsound steps in it (which is likelier to happen with some of the lighter translations)

improve sort inference in Metis proofs -- in some rare cases Metis steals Isabelle's variable names and the sorts must then be inferred as well

removed needless function that duplicated standard functionality, with a little unnecessary twist

removed more dead code

be a bit more liberal with respect to the universal sort -- it sometimes help

renamed "untyped_aconv" to distinguish it clearly from the standard "aconv_untyped"

merged

avoid duplicate facts, which confuse the minimizer output

pass Metis facts and negated conjecture as facts, with (almost) correctly set localities, so that the correct encoding is used for nonmonotonic occurrences of infinite types

restore comment about subtle issue