Implement uninstall tool command.

[peterhuene]

This commit implements the uninstall tool command.

The uninstall tool command is responsible for uninstalling global tools that
are installed with the install tool command.

This commit heavily refactors the ToolPackage and ShellShim namespaces to
better support the operations required for the uninstall command.

Several string resources have been updated to be more informative or to correct
oddly structured sentences.

This commit also fixes --version on the install command not supporting ranges
and wildcards.

Fixes #8549.

Issue #8485 is partially fixed by this commit (--prerelease is not yet
implemented).

[peterhuene]

So two tests appear to be failing on the localized build because (I assume) it uses the existing translation for these resource strings; given that the arguments to the formatted strings have decreased, using the existing translation would result in this failure (i.e. we are now passing in a single argument when an existing translated string would expect {1} to work).

When modifying a string with an existing translation such that the number of formatting arguments decreases, is it best to just rename the string resource, thereby ensuring the localized build tests with the correct number of expected arguments?

[nguerrera]

Yes, rename if fewer args are passed.

[peterhuene]

Looking into the RHEL failure.

[wli3]

The following do not need to address in this PR

I just realize a lot of stuff we don't have to write mock, The current mock has a lot of duplicated code with production, which is a sign that they can be extracted out.

Since the only place we cannot easily test is the part with nuget(the original obtainer part) and shim creation, everything else can be test against these mocks (may be obtainer mock need to write the xml file, since the xml file is considered a contract, not implementation detail anymore). In a sense our adapter layer is too high, a lot of them can be considered core logic.

The logic of transaction, folder location, rename folder, delete assets, can all be test against package obatiner mock, shim creation mock, and file system mock. The above logic can be tests all in memory. In this way, we don't have to write mock for package store, duplicate transaction logic in mock. To test InstallCommand, we can test again concrete transaction logic but just cut off the layer that deal with file IO.

Maybe we can code list logic as such.

[peterhuene]

The RHEL failure is legitimate and appears to be a race. I'll try to investigate and fix tomorrow.

[peterhuene]

I've pushed up an amendment that prevents the transaction manager from creating a timer to timeout the transactions. This address the inconsistent RHEL failure and also the issue of very long tool package downloads as they would have been aborted as a result.