Stop dumping GAM to stdout when reading >1k BED or GFF records

[adamnovak]

Changelog Entry

To be copied to the draft changelog by merger:

• When reading more than 1000 BED or GFF records, vg will no longer dump the first records to standard output and forget about them

Description

In ae44fca @jmonlong improved memory usage when converting large BED and GFF files to GAM, by adding points at which we flush the output buffer. But when I merged #4248 I didn't notice that that buffer isn't always an output buffer and is sometimes an internal collection of annotations to work with later.

So whenever we tried to load and keep a lot of annotations, we would dump them out in GAM chunks of 1000 records each and throw them away, whenever we hit 1000 loaded records.

This fixes that.

[jmonlong]

Sorry about the bug I introduced @adamnovak. I didn't notice that region-parsing function was used somewhere else...

I noticed the issue this week when working to make vg annotate output GAF. In #4364 I think we get both the fix and GAF output feature. If it looks good to you, could we merge this one (or add a GAF output mode to yours)?