Remove requirement of SparkR

[wjsi]

As R users are not always using SparkR, it is better to activate the Spark DataFrame serializer only when SparkR is installed.

BTW, check_packages() in declarativewidgets/kernel-r/declarativewidgets/R/serializers.r always returns TRUE. It is also fixed in this PR.

[poplav]

@siyuwj

That is a good point. I see a few things that can be done:

• In load_serializers we should first check the serializer has it's dependencies via check_packages. This is a bit cleaner than the initial commit for this PR and is what the python side does here. We should also add logging that the package failed to load and the serializer was not inited correctly.
• The tests directly use sparkR also, so those should be separated out and logged that sparkR type tests were skipped due to unmet dependencies.

[wjsi]

@poplav I added check_packages in register_serializer. Spark tests are skipped when SparkR is not installed.

[poplav]

@siyuwj I think your changes look good, we are having issues with tests at the moment so bear with us regarding timing of merging.

[poplav]

@siyuwj Thanks again for your contribution. We recently finished a revamp of our testing/build process at #532. Can you rebase this with master and then we'll merge it!