Fieldlists: Refactor command line arg processing, part 1

[jondegenhardt]

This PR is a follow-on to the named field PR series start with PR #284.

This PR addresses one of the main issues with using named fields - The header line must be read before command line arguments using field names can processed. If the tool is in the later stage of a unix command pipeline, it might be a while before the tool receives data and reads the header line. An error in the command line arguments will terminate the operation. If the data is large, this could occur a decent period after starting the operation.

This PR addresses this in a couple of ways. First, command line arguments not needing access to the header line are processed first, prior to reading the header. Errors for invalid command line arguments are output immediately. If header lines are not being processed (no --H|headers), then this includes field lists, as they must be numeric. Second, header lines are output immediately, prior to processing other input. This has the effect of passing header lines down the Unix command pipeline. Command line argument handling can occur much earlier, without a lengthy time delay.

This PR implements this for: tsv-select, tsv-filter, tsv-sample, and tsv-join. Other tools will be updated in a subsequent PR.

This is a step towards enhancement request #25.