Fix/improve Factor Observation JSON serialization

[Affie]

Notes:

• Currently, FactorDFG is a concrete type, that makes serialization and multilanguage support easy.
• FactorDFG exists purely for serialization and multilanguage support.
• Ideally the packed observation should be [de]serialized directly to JSON and not include the extra string step with manual type creation. This complicates serialization but may be easier in JSON.jl v1
• First wait for [BREAKING] Package rewrite for a 1.0 release JuliaIO/JSON.jl#374 and maybe upgrade all serialization to JSON.jl in one go.
• Also decide if observation is a type parameter or abstract type in FactorDFG -- looks like we don't want it as a JSON string any more. Other language SDKs and serialization will likely influence this.
  • We get closer to dropping the difference between FactorDFG and FactorCompute if we go with type parameter.
• We may want to be able to use the observation independent from the factor, ie. getObservation, if true, <: Observations will likely need a type field in the JSON string and not only depend on the factor type. xref Restore for variables and factors the module prefix #1140 (comment)
• We can/should consolidate FactorDFG and FactroCompute (as FactorDFG)

[dehann]

observation should be [de]serialized directly to JSON and not include the extra string step with manual type creation.

Agree

First wait for JSON.jl 1.0

Agree, JSON.jl would be nice. Hopefully that drops soon...

[Affie]

Question from #1119

"should we parameterize FactorDFG{T}, whats going on in SDK.rs"

[Affie]

See https://github.com/JuliaRobotics/DistributedFactorGraphs.jl/blob/develop/src/serialization/PackedSerialization.jl for the leading idea and #992 (comment)