Prototype: Support for dynamic loading of provider definitions

[asilverman]

Overview

Adds support for loading the 'az' provider dynamically sourced from an OCI artifact registry. The artifact must be restored (downloaded to the appropriate location) manually under $BicepCacheRootDir/br/mcr.microsoft.com/bicep$providers$az/${providerVersion} which is the target location in the Bicep cache for the az provider.

The work to restore the provider data to the cache will be handled in a separate PR for convenience so that reviewing PRs is easier.

To enable the feature DynamicTypeLoadingEnabled and must be set to true in bicepconfig.json:

{
    "experimentalFeaturesEnabled": {
        "extensibility": true,
        "dynamicTypeLoadingEnabled": true,
    },
    "cacheRootDirectory": "~/.bicep",
}

Changes

• Adds a new experimental feature flag DynamicTypeLoadingEnabled
• Adds a new factory AzResourceTypeLoader that is handling the concern of deciding the provider loader to use based on the feature flag and presence of an import declaration syntax in the Bicep file being processed.
• Serializes the provider version by inspecting the ImportDeclarationSyntax vs using a hardcoded value

Fixes #10662

Contributing a feature

• I have opened a new issue for the proposal, or commented on an existing one, and ensured that the Bicep maintainers are good with the design of the feature being implemented
• I have included "Fixes #{issue_number}" in the PR description, so GitHub can link to the issue and close it when the PR is merged
• I have appropriate test coverage of my new feature

Microsoft Reviewers: Open in CodeFlow

[asilverman]

I think this is laudable here but at the same time unsustainable @majastrz, @anthony-c-martin should we add a ticket to have a framework to test performance so we can apply consistently the necessary checks and also prevent unintended regressions?

[StephenWeatherford]

Feel free to get a second opinion on the syntax tree in type provider code issue.

Stephens concerns were assuaged by manually testing performance