Adding overload for hpx::init/hpx::start for use with resource partitioner

[hkaiser]

• this helps resolving linker errors on Windows if the main entry point to HPX is in a shared library
• this also cleans up the API as it makes the use of the resource partitioner explicit

Note that this is a breaking API as the use of the no-argument versions of init and start is now deprecated when using an explicit resource partitioner.

[msimberg]

Thanks @hkaiser for starting this. @aurianer has another PR almost ready to go which actually passes a callback to init/start to create thread pools etc. with the resource partitioner. This PR still allows initializing the rp explicitly outside of init/start and I was actually hoping to disallow that, but we could go this way as well. The critical change that I wanted to have is to not have the rp grab things like parcelport configs directly but have them passed to it by init/start. Is this for a specific use case of yours or related to the discussions we had earlier on cleaning up use of the rp with init/start?

[hkaiser]

Thanks @msimberg. This change is to avoid a linker error on Windows under certain circumstances. The problem I'm trying to solve is that in cases were the resource-manager is initialized explicitly (with an explicit HPX-main function) the no-argument overload of hpx::init still passes hpx_main as a symbol to the runtime, causing linker errors in applications. This new overload exposes a separate entry point function that works around the issue.

This PR also makes passing the resource-partitioner to the runtime explicit. I never liked the implicit 'pseudo-global' resource partitioner that was snug in behind the scenes.

I have no issue to achieve the two goals in any different way, however.

[msimberg]

Thanks @hkaiser. I suggest we go with this first (and if you can add a small test demonstrating the linking problems that would be great). We'll adapt @aurianer's changes to fit this.

[msimberg]

Looks good to me.

[aurianer]

Thanks !

[msimberg]

I already forgot about the test when I merged this. Let's figure out a test on IRC.