"Strict mode" the same behavior as in ruby's

[msangel]

As a result of detected problems in https://github.com/bkiers/Liqp/issues/267 todos:

• [X] error_mode setting
• [X] remove used alternative in particular cases
• [x] apply in the parser for the same incorrect expression evaluating in {{ output }} tags as for https://github.com/Shopify/liquid/issues/1102 according to existing logic (lax vs strict)
  • [x] apply incorrect expression evaluating
  • [ ] lax vs strict
  • [x] parser differentiation in vocabulary
  • [x] parser implementation detect garbage and record error (lax vs warn)
  • [x] Where filter still rely on exact flavour
  • [ ] more test coverage!
• [x] Flavor.filters vs ParserSettings.filters differentiation, make more use of new Filters class(immutable)
• [x] get rid of static context in Filter
• [x] get rid of static context in Insertion
• [x] ProtectionSettings is not immutable right now as the counter live in it, counter should be moved in context (probably as registry)
• [x] in LIPQ flavor enable expected behavior (expressions are evaluated)
• [x] undo string-to-digit conversion as for Gt, GtEq, Lt and LtEq and allow it only for the same family type(numbers, dates, strings)
• [ ] more unit tests

[kohlschuetter]

This looks good overall. Even if not all to-dos are met, tests pass. The earlier we get this in the better :) 👍

Please see my additional commit below for a regression I just found after clearing up all the other warnings.