Use a single FacterDB.get_facts call

The major cost is FacterDB.get_facts. Instead of calling it once for
every OS version this instead calls it once with a complex filter. It
then uses JGrep to simulate another smaller DB and futher query that.

In my non-scientific test this reduces the load time of puppet-example
from 3.5 to 2.5 seconds.

lib/rspec-puppet-facts.rb

@@ -118,11 +118,21 @@ def on_supported_os_implementation(opts = {})
     loose_requirement = RspecPuppetFacts::facter_version_to_loose_requirement(facterversion)
     received_facts = []
 
+    db = FacterDB.get_facts(filter)
+    unless db.any?
+      RspecPuppetFacts.warning "No facts were found in the FacterDB for: #{filter.inspect}"
+      return {}
+    end
+
     # FacterDB may have newer versions of facter data for which it contains a subset of all possible
     # facter data (see FacterDB 0.5.2 for Facter releases 3.8 and 3.9). In this situation we need to
     # cycle through and downgrade Facter versions per platform type until we find matching Facter data.
     filter.each do |filter_spec|
-      versions = FacterDB.get_facts(filter_spec).to_h { |facts| [Gem::Version.new(facts[:facterversion]), facts] }
+      versions = JGrep.jgrep(db.to_json, FacterDB.generate_filter_str(filter_spec)) \
+        .to_h do |facts|
+          result = facts.to_h { |k, v| [k.to_sym, v] }
+          [Gem::Version.new(result[:facterversion]), result]
+        end
 
       version, facts = versions.select { |v, _f| strict_requirement =~ v }.max_by { |v, _f| v }