When a device was not used while filter initialization use a default
one value

This is obsolete and produces invalid values while query

The attributes use b-tree internal structure what is x86-specific
implemented. Also b-tree lookup is slower that direct table lookups.

This is a preparation commit introducing an attribute compilation
primitives. The commit introduces a some sort of polymorphism required
to get rid of filter library hard-coded filter rule format and packet
handling.

Filter logic implies the following logic:
 - each attribute definition area is split into regions with unique set of
   rule set corresponding to the region
 - each rule gets a list of region identifiers for each attribute
 - combine pairs of attributes into sets of next level identifiers
 - the last stage resolves corresponding rule index

So in case of 5 attributes compilation schema for a rule looks like
a0, a1, a2, a3, a4 -> sets of range identifiers matching to the rule
produce a5 = a0 * a1, a6 = a2 * a3, a7 = a4 * a5 where ai * aj if s full
join of containing values
after then we set rule index for each pair in set of a6 * a7.

The logic consist of three stages
 - initialize all the attributes and touch regions for all rules. The stage
   result in enumerate all regions with values defining of unique rule set
   identifier. Also we collect region values for each rule into registries.
 - merge pairs of values into a new ones
 - after only two set of identifiers remained - the last one merge will set
   a rule index into corresponding table.