Liana-py

Changes
- Change the order of non-expressed genes and empty droplet filtering.
I now appropriately filter cells in the `sce` object *after* limiting the gene
universe to ligands and receptors in the resource.

Minor changes
- Appropriately pass `verbose` to `.filter_sce`

- Silence expected warning in `cellchat_formatDB`

New Implementations

- We now provide a tutorial dedicated to `orthology conversion` of the resources in LIANA

Minor Changes
LIANA will now check if:
- There is enough of a gene intesect between the resource and the data (i.e. if they are both for human)

- There are negative counts

- An assay with normalized counts was been provided (i.e. if the data/logcounts slot is scaled).

- LIANA will now convert non-sparse matrices to sparse.

Changes

- For `logFC_mean`, rather than normalizing the counts by library size, I instead inverse log the counts and use those to calculate log2FC. This is to preserve any prior correction of the counts,
i.e. mainly for consistency with the rest of the methods.

Minor Changes

- Flipped `x` and `y` axes dotplot according to feedback.

- Added `.default_fun` parameter to `generate_orthologues`

- Minor code clean up

- Added `examples` to main exported functions docs

- Removed several low quality interactions from the `Consensus` resource

- CellChat will now work with the simplified format of intercell (i.e. Consensus resource).

- CellChat and Squidpy should now be called using `call_cellchat` and `call_squidpy`, respectively.

Changes
- Add `prod_weight` to NATMI's score. This is the weight that both Connectome and
NATMI suggest for between-condition comparisons. Add NATMI to the housekeeping
aggregate ranking.

- Enable weighing of interactions by cell pairs (using a DF in which each
cell pair has an assigned weight). This would typically be done by spatial
constraints, etc. These weights can also be used to mask any cell-pair interactions
which are not relevant (by assigning weights of 0). This currently assumes that
the weights would be between 0 to 1 - to be extended. Tutorial on this /w appropriate
spatial weight generation to be written.


Minor Changes
- By default, the base for logFC will now be automatically assigned depending
on the object passed to LIANA, i.e. `.antilog1m` for SCE will use 2 as base,
and Euler's number for Seurat. One could also pass the base they wish to use
via `liana_wrap`.
- Automate website deployment to gh-pages and run R checks on push.

Changes
- Changed the aggregation columns of `liana_aggregate`, as in some cases
methods would assign different subunits as the minimum, which results in
redundancies for the same complex. As such, `liana_aggregate` will now
return only the complex columns, nevertheless, the methods will still return
both the minimum (lowest expressed subunit) and it's corresponding complex.

- `base` used to calculate logFC (via `get_log2FC`) can now be passed as
a parameter to `liana_wrap` via `liana_pipe.params`.
Passing `NaN` to base would result in log2FC calculation using the raw counts
without any pre-processing (e.g. no batch correction, etc).

The base is by default set to 2, assuming that log2 transformation is performed
following library size normalization, and thus preserving the normalization,
while reverting back to ~counts.

Minor Changes
- Extended the heatmap to allow filtering down to certain cell types.
- Removed redundant/leftover code

New implementations
- A chord plot for interaction frequncies was included.

Changes

- Changed the way ties are handles in liana_aggregate. Namely, I previously
assigned the minimum rank, but this resulted in ties getting lower p-values
than they should, particularly for scores with a lot of ties (e.g. CPDB p-value).

- Fixed an issue where some subunits of 0 `expr_prop` would not get filtered.
This was observed due to previous changes to `.filt_liana_pipe` in 0.1.6, where
some subunits were filtered before `recomplexifying`.

- Fixed an issue in which some NATMI complexes would be missing due to `recomplexify`
being done on both .expr and .sum columns. These are now seperated into `columns`, which
are the ones for which I account for complexes, and `add_cols`, the ones that
are additional - no need to account for complexes (e.g. also `global_mean`).

Minor Changes
- I now refer to [SCPubr](https://enblacar.github.io/SCpubr-book/index.html) in liana's tutorial.
- Throw exception for NAs in cell idents
` Remove duplicated rows from orthologous resource

Minor changes

- I now refer to SCpubr in the tutorial for more advanced plots.