Allow interfaces to be declared and implemented within intrinsics

[gusty]

This was discussed as part of fsharp/fslang-suggestions#691

[gusty]

@KevinRansom I don't understand why I'm still getting this error stating that there's a difference in neg46 when I already removed the relevant part from the neg46.bsl

[cartermp]

@gusty if you go here (for some reason not working on my mac laptop) and click on fsharpQA logs you can see detailed failures: https://dev.azure.com/dnceng/public/_build/results?buildId=1030986&view=artifacts&pathAsName=false&type=publishedArtifacts

[gusty]

So far here we're removing the limitation.

Another PR could adjust the warnings as suggested by @dsyme and remove it completely when the containing module is a rec one.

Please consider including this in the next preview as we need this in our projects to separate the business logic (at least in the same file at the top).

[Happypig375]

Bump @dsyme

[dsyme]

@gusty The PR doesn't match what's approved?

The limitation ...[is to be]....turned into a warning when "namespace rec" is not used, and removed altogether when it is.

[gusty]

@dsyme as I said, it doesn't remove the warning in the rec case, so yes it doesn't match 100% but it fails (warns) on the safe side.

Anyway, I can have another look at it and try to suppress the warning in those cases, last time I checked I was unable to figure out where is that module information available. Any help would be much appreciated.

[dsyme]

@gusty The PR doesn't add any tests showing what is now allowed

To be clearer, the approved suggestion is

1. The warning message

   a.fs(8,14): warning FS0069: Interface implementations in augmentations are now deprecated. Interface implementations should be given on the initial declaration of a type
   

   gets changed to

   a.fs(8,14): warning FS0069:  Interface implementations should normally be given on the initial declaration of a type. Interface implementations in augmentations may lead to accessing static bindings before they are initialized, though only if the interface implementation is invoked during initialization of the static data, and in turn access the static data.  You may remove this warning using #nowarn "69" if you have checked this is not the case."
   

   I initially said it should be removed but I think it's more like warning 40 given its about about initialization soundness - something that can be optionally suppressed.

2. The warning message is suppressed if namesapce rec is used

It's ok to just do part (1), since it;s only a warning message. Please also add tests for this as there will be very few or none in the codebase.

[T-Gro]

@gusty Hi, fine if I take this over ?

[gusty]

@T-Gro Sure, go ahead.

It's basically finished, just need to remove the warning in the mentioned case.
I didn't remove it as it requires some information about the module which is not directly available so have to figure out where can I pull it from.

[T-Gro]

@gusty ; @dsyme ; @vzarytovskii

Checks passing now, ready to be reviewed.

[gusty]

Looks good to me, I can't approve as I'm considered as author. Thanks for taking care of this.

[dsyme]

Just a couple of minor nits. Looks great!

I'll mark as "comment" to not block approval - consider it approved besides these minor nits

[dsyme]

@T-Gro I merged main into this

[dsyme]

@vzarytovskii This could get mentioned in F# 7 release announcement, perhaps also a very brief RFC that the limitation has been lifted.