clustergen: Improvements for generation with templates

[fischeti]

This PR aims to improve the auto-generation of sources in several ways:

Cacheable regions

It is now possible to define external (cacheable) memory regions in the configuration file:

external_addr_regions: [
        {
            name: "dram"
            address: 0x8000_0000,
            length: 0x8000_0000,
            cacheable: true
        },
        ...
    ],

Previously, that was not possible. Instead, the intended way was to write a python script that imports the SnitchCluster class and adds additional information such as cacheable regions. For instance, for the testbench of the snitch cluster itself, there exists a SnitchClusterTb class which defines a simple environment with a (cacheable) DRAM. However, writting a python script for integrating the snitch cluster is unnecessary cumbersome, which is why this was removed in favor of external_addr_region.

Arbitrary templates

Instead of having different flags for the generation of different kind of sources, the clustergen now just validates a configuration file and extends it with additional information, from which arbitrary sources can be generated by providing a template with --template.

Migrate to JSON5

The configuration files were migrated from HJSON to JSON5, which allows hexadecimal input, as well as comments

[colluca]

Thanks for the contribution @fischeti!

Just a few comments from my side, otherwise looks good to me.

[colluca]

I pushed a couple of commits to streamline the clustergen tool even more. Summary below. @fischeti if you can give them a quick review, then for me it's ready to merge.

1. It seems the memories.json file is not currently used anywhere in this repo, nor in Occamy. I temporarily removed it, to make the interface to the tool more consistent. In the future we can use a template to generate this file as well, if needed.

2. Enable overriding the output file name, by passing a full path to the desired output file, instead of just the output directory.

[fischeti]

Looks good to me. The memory.json was previously used to generate the technology-specific tc_sram wrapper. But I don't think it is needed anymore