Preserve yaml key order, render tabular file formats

[rcannood]

Two bugs that make a rendered task README wrong or unstable.

Key order

deep_merge() collected the merged keys in a set(). Python randomises string hashing per process, so every yaml read through read_nested_yaml() came back with its keys in a different order on every run. The author info columns of a task README are built by iterating those keys, so the README changed on each render:

PYTHONHASHSEED=0  | Name | Roles | Linkedin | Twitter | Orcid | Email | Github |
PYTHONHASHSEED=1  | Name | Roles | Orcid | Email | Twitter | Github | Linkedin |
PYTHONHASHSEED=2  | Name | Roles | Twitter | Linkedin | Github | Orcid | Email |

Every task repo got a spurious README diff each time create_task_readme ran. The keys now keep the order of the first object followed by the ones only found in the second, which is what the R deep_merge() has always done. With the fix the rendered README is byte-identical across seeds.

Tabular file formats

read_file_format() accepts tabular as a format type, but both renderers only matched csv, tsv and parquet -- so a tabular spec produced rows that were never rendered, leaving an empty Format and Data structure section.

Tests

• deep_merge(): merging, recursion, list appending and key order
• read_nested_yaml(): keys come back in the order they appear in the yaml, checked against a plain yaml.safe_load() of the same file so the test does not go stale
• render_file_format(): parametrised over all four tabular types

Both tests fail on main (the key order one on every hash seed).