Graphannis

Fixed

- Fixed wrong result order for non-token searches.
- Estimation for negated regex was extremely off when the regex could possibly
match all values. This caused problematic query plans including those with
nested loop joins and long execution times.
- Better estimation of result sizes for regular expressions with multiple
prefixes.
- Fix compilation issues in Rust projects that use the 2021 Rust edition.
https://github.com/lalrpop/lalrpop/issues/650
- Faster subgraph generation for `subgraph` queries with context. The previous
implementation used an AQL query that got quite complex over time and was
difficult to execute. The new implemenation directly implements the logic
using iterators. It also sorts the nodes in the iterator by the order of the
node in the text.

Added

- Add edges to the special `Ordering/annis/datasource-gap` between the last and
first token of context regions in `subgraph` when the returned context regions
do not overlap. This allows sorting the context regions that belong to the
same data source but are not connected by ordinary `Ordering/annis/` edges.

Fixed

- Use external sorting for match results to avoid out of memory errors for large
results.

Fixed

- For subgraph queries with segmentation, the left and right context was
switched.

Added

- Allow to configure the expected display order of (sub)-corpus meta annotations
using the `corpus_annotation_order` field in the view configuration.

Changed

- Update to graphANNIS 2.1.0 This does change the graphANNIS C-API (see
<https://github.com/korpling/graphANNIS/blob/main/CHANGELOG.md#200---2022-03-29>)
but does not actually the exported functions of the Python API.

Fixed

- Fix subgraph generation when a segmentation was defined as context and the
match includes a token that is not covered by a segmentation node (there are
gaps in the segmentation). This is achieved by explicitly searching for all
token between the first and last matched segment and produces a more complex
query than before. Because token where missing from the graph, it could appear
in ANNIS that there are gaps in the data and that the token order is
incorrect.