sumtype: Work around bugs in return type inference

sumtype: Use non-packaged import for std.format ...

std.format was split into modules in v2.097, but we want to be compatible
with older versions of the frontend, so use the package as a whole.

sumtype/dyaml: Work around various compatibility issues with v2.097

dyaml: Fix unittest compatibility with v2.099

Do not use Phobos' sumtype, use dyaml's versioned one ...

So that we can compile with older versions of the frontend,
currently v2.097 instead of v2.098.

Return null instead of empty array

Fix zip symlink problem.

Dub: Deprecate changing the rootPath ...

It's an essential part of the setup, and too many properties are derived
from it that we don't want to let the library user the ability to mess with it.

Commandline: Use inheritance, not a boolean, to control Dub initialization ...

The boolean was a breach of abstraction, as it was introduced
for a single sub-command. Instead, use inheritance, which
allows to keep that logic compartimentalized to the subcommand itself.

Copy all dynamicLib deps extra files for root packages with `none` targetType ...

To support 'framework' use cases - a dummy root package depending on a bunch
of dynamicLibs, where all extra files are to be copied to the output dir.

Copy extra files of (direct & indirect) dynamicLibrary deps to root package output dir only

Windows: Copy .lib files from DLL deps only ...

One aspect of issue #2377 - don't copy static libs. And only copy
import .lib & .exp if the dependee is a DLL too.

Use dscanner-config for `dub lint` ...

Keep all original commandline flags in case we want support them in
the future.

Fixes #1940

Support --temp-build also in `dub build` (#2428)

Fix package scanning order. ...

Packages registered using add-path need to come before cached packages.

Avoid throw expression - fixes compilation on LDC 1.30.0.

PackMan: Provide an overload of storeFetchedPackage that doesn't expose Location ...

This way we keep the internals of PackageManager within the PackageManager,
instead of having external classes be aware of the paths we use.

fix coloring for dub upgrade -s

PackageManager: Only iterate on internal fromPath array, not everything ...

getOrLoadPackage is only used for path-based dependencies,
which are only stored in m_internal.fromPath.
We can thus reduce the iteration by only using fromPath.
If someone had a path-based dependency that pointed to a version-based package,
it would lead to this package being loaded twice, but that's not really a concern,
as it would not lead to problem either.

Remove PlacementLocation information from Dub.fetch ...

Let it rely solely on the PackageManager instead.

PackMan: Deprecate overload leaking location path ...

Currently, a lot of the internals of the PackageManager are leaking.
The getPackage overload accepting both a version and a NativePath
is an example of such a leak, and looking at the usage sites
show that things can be simplified by taking a PlacementLocation,
which is a higher level representation of the path.

Move PackageManager.getPackagePath to Location ...

It is more suited as a Location method, as it returns a path based on PackagePath.
This also allows us to reduce the leak of information about Location
happening in dub.

Deprecate PackageManager.completeSearchPath ...

It is no longer used within the PackageManager and leaks quite a bit of the internals,
instead users should have functionalities exposed via the PackageManager's API.

Dub: Do not re-load already loaded tool Package ...

Instead of giving the path to the Package we just loaded,
just give it the Package we just loaded.

Deprecate an unused PackageManager.getPackage ovrld ...

This overload is not used anywhere in `dub`, and its use case is not clear.
Path where packages are stored should ideally be only known to `PackageManager`,
and the API should expose `PlacementLocation` instead of bare `NativePath`.
With this in mind, this overload doesn't fit the picture.

Dub: Remove redundant variable m_projectPath

Deprecate PackageManager.getFirstPackage ...

This function returns the 'first' package, which is very arbitrary.
Instead, always return the highest matching package for stable results.

PackageManager: Simplify removeLocalPackage by using algorithm ...

Instead of doing (potentially) multiple append operations of two slices,
just do a filter + array which is much easier to read and prove correct.

PackageManager: Remove m_disableDefaultSearchPath, use array instead ...

The boolean is redundant, as the array will always be null / empty
if it's true, and have 3 entries otherwise.

Trivial: Improve PackageManager.m_repositories documentation