speed up MaSh queries

use better score function, based on previous evaluation (cf. Deduct 2011 slides)

attempt at meshing according to more meaningful factors

don't include hidden facts in relevance filter + tweak MaSh learning

removed debugging output

removed expensive HO check in MaSh

speed up tautology/metaness check

optimized MaSh output by chunking it

fixed MaSh state load code so it works even if the facts are read in disorder

learn from minimized ATP proofs

improved meshing of MaSh and Meng--Paulson if some MaSh suggestions are cut-off (the common case)

use async manager to manage MaSh learners to make sure they get killed cleanly

more consolidation of MaSh code

removed lie

drastic overhaul of MaSh data structures + fixed a few performance issues

fixed order of accessibles + other tweaks to MaSh

added option to control which fact filter is used

mesh facts by taking into consideration whether a fact is known to MeSh

implemented meshing of Iter and MaSh results

implemented MaSh QUERY operation

refactored MaSh ADD code so it can be used for SUGGEST as well

implemented low-level MaSh ADD operation

make tracing an option

cleaner handling of metacharacters + freshness of one-off facts

better zipping of MaSh facts

implemented MaSh learn theory function

more work on MaSh

improved MaSh string escaping and make more operations string-based

more implementation work on MaSh

started implementing MaSh client-side I/O