* Add env feature to clap and cargo update

* Dead Code: WIP CLI changes

* Dead Code: CLI feature parity

Although see clap-rs/clap#5020

* Note about not (yet) using infer_subcommands

See clap-rs/clap#5021

* Use clap/wrap_help

Although see clap-rs/clap#5022

* WIP: Strip out previous CLI argument parser setup

[skip ci]

* WIP: Normalise arguments for caller

[skip ci]

* Don't check for file-ness in the CLI argument parser

* File and directory canonicalisation for the argument parser

* Expose CLI types so they can be used downstream

* Update CHANGELOG

* Update README

* Update script that checks that the README correlates with --help

* Make README/CLI checking script a bit clearer

* Updated usage with new collation modes

* Documented configuration collation

* Update example call sites to new CLI

* Added a migration guide for the CLI changes to the CHANGELOG

* Update TTY playground to use new interface

* Update pre-commit hook to use new interface

* Implement Display for Configuration: pretty-print TOML

* Comment out old code, stub out new interface implementation

* Additional collation mode

* Some light refactoring

* More refactoring...

* Make the config collation independent of a specified config file

As there are several potential sources of configuration, the collation
mode is relevant whether a invocation-specific configuration file is
specified, or not. This has the cheery consequence of sidestepping
the issue described in clap-rs/clap#5020.

* Add useful annotations to the computed TOML output

* Refactor 'revise' collation mode from original

* Remove unnecessary trailing \n from TOML output

* Syntax tweak

* Add tests for TOML collation

* Refactor: Update and use (non-differentiating) collation API

* Got `merge` collation working

Includes slight refactoring of Helix's TOML merging function, largely
for my sake while I was trying to understand it!

* Some sugar for your tea?

* Note about not using infer_subcommands

* --tolerate-parsing-errors doesn't make sense for visualisation

* Separate out input source types so we can create a unified interface

* Fallback to the given output path if canonicalisation fails

Resolves #588

* We're going to need an InputFile, too

* WIP: InputFile type

* Correct blunder regarding --query

Note that clap doesn't support (--foo [--bar] | --quux) groups very
cleanly; it was a bit of a hack to get this to work, with the result
being the error text being a bit off when an illegal combination is
attempted. I've attempted to compensate for this by making the long help
text quite explicit.

Also updated clap, which contains the fix for clap-rs/clap#5022

* Missed change to README

* Add note RE clap-rs/clap#4707 workaround

* Machinery to unify inputs + downstream use to reimplement visualisation

* Don't flatten-away errors

* Don't open the input file until we need to read from it

* Into InputFrom should be from &AtLeastOneInput

* Add logging

* Moar logging!1!!

* Beginnings of thread-safe language definition caching

* Expose the raw values and implement Display

* Thread-safe cache of language definitions

* WIP: Formatting

Formatting is now functional, but there remain implementation details to
sort out. Bufferin, caching and error handling, in particular...

* Add buffered reading and writing

* Make the cache operational, if not suboptimal :(

* Better error handling

* topiary fmt tests

* Fix test task in CI workflow

* Try increasing web test timeout to 30s

* Make sure GNU sed is available, even on macOS

* CLI tests for visualisation

* CLI test for configuration output

* Expose logging verbosity through CLI flag

* Updated documentation

* Lock the cache with Tokio's Mutex

* Use the entry API as it's a bit cleaner

* Remove FIXME comment

* Initialise the logger as early as possible

NOTE: nixfmt also had its way with the source; hence the unrelated diffs

This reverts commit 5510406.