Co-authored-by: annes <76728727+annesnour03@users.noreply.github.com>

* Added current-state snapshots

Capturing the current state of transformations so we can see how the
upcoming changes affect them.

* Add `<SEPARATOR>` support

When `importOrderSeparation` is enabled, users can further control
separation placement with the `<SEPARATOR>` keyword.

Switch to ESM

* Require node >= 20.x

* Add checking 24.x

I was still hitting issues with jest and trying to get full ESM support, so I decided to switch to vitest. This seems to be working and snapshots are generated essentially the same.

Switch to vitest for better ESM support

Summary:
This logic can hide breakages and cause tests to pass when the
underlying logic is actually broken. I encountered these hidden
breakages while working on some further feature changes, and figured
it was appropriate to fix this first.

As all existing tests pass without the try/catch, this change is
effectively a no-op.

Test Plan:
`yarn install && yarn run test ./tests/`

[Test] Stop ignoring exceptions thrown in snapshot tests

Summary:
This change replaces the existing RegExp replace() logic with
concatenated string slices. This avoids reallocation the result
string for each node replacement, replacing it with string slice
operations (which are implemented as O(1) string views within v8) and
a single .join(''), which can be optimized by the runtime to a single
allocation.

This probably won't make a noticable difference, but the change also
simplifies some further feature work I am attempting to add.

Test Plan:
`yarn install && yarn run test --all`

[Perf] Improve performance of removeNodesFromOriginalCode()

Summary:
This commit changes how the sorted imports are combined with the
original source.

Prior to this commit, all ImportDeclaration nodes and their leading
comments, plus any InterpreterDirective and Directive nodes, were
extracted from the original code and re-rendered using babel. The
rendered nodes were then concatenated with the original source with
those nodes removed to produce the updated source.

This approach safely protected against functional changes, but
removed newlines around comments near the beginning of the file when
the first node of the original source was an ImportDeclaration, as
babel does not preserve whitespace when rendering content.

If a user has configured a plugin that attempts to manage comments
and/or whitespace near the top of the file, such as auto-inserting a
license header (as I am trying to do), this results in conflicts /
formatting churn.

This commit does not directly resolve this incompatibility, however
it better prepares the codebase for a plugin option to be added that
can resolve the issue.

Test Plan:
`yarn install && yarn run test --all`

Note that one snapshot was changed by this commit where a newline was
changed, acting as an effective example of how the original approach
could affect whitespace in the re-rendered portion of the file.

[Refactor] Stop rerendering directives, inject imports instead

Summary:
Prior to this commit, isSortImportsIgnored() checked comments
associated with the extracted ImportDirectives, and only if the
comment started on line 1. This could result in a failure to suppress
sorting if the comment was not next to an ImportDirective (never
considered), the first import directive were embedded later in the
file (line mismatch), or directives/shebangs were used (Line 1 is
unavailable).

With this change, the line restriction is removed, and all comments
from the beginning of the file and the first statement are checked.
This ensures better coverage, especially with the
importOrderIgnoreHeaderComments feature stacked on this commit.

Test Plan:
`yarn install && yarn run test --all`

[Feature] Expand detection for `sort-imports-ignore`

Add <BUILTIN_MODULES> option for Node.js builtin modules

ignore files

This reverts commit 5b1b267.