create more efficient model for specifying regions of alignment to be exported as flat file

The current paradigm for writing an alignment file from Jalview is to create a new set of SequenceI + AlignmentAnnotation objects wholly containing the sequences + annotation to be output. This happens either via AlignmentView objects or alignment objects plus strings specifying 'visible' regions of sequences.

Ideally, parsers should be able to take a lightweight set of references specifying the regions of an alignment to be output. Currently, parsers are hardwired to work with a vector of SequenceI objects plus annotation - refactoring is needed to allow proxy classes to be used such as the AlignmentView/SequenceCigar objects, or even the AlignViewport object itself (if thread safety allows it), to avoid duplicating/triplicating the memory needed when exporting alignment data.