name should be package name, not store path name

[roberth]

Issue description

The name part of a store path an extra suffix that's human readable, unlike the rest of the path.
We use it to convey the package name, and usually the version as well.

However, because we tend to include the version as well, this means that in general, the package name is not equal to the store path name.

Currently the .name attribute of the package attribute set contains the store path name. This is surprising.

Goal

• Make pkg.name contain the package name and nothing else
• Make pname a mkDerivation implementation detail
• Stretch goal: remove the need for pname; allow name and version to be specified, in such a way that they match the package attributes and the version gets added to the store path.
• Clean up the package attribute set ahead of RFC 92 packages and multi-derivation packages (e.g. split output build)

Refs

• Found again in
  • __cleanAttrs: Clean packages that don't expose all internals #217243
• Corresponds to drv-parts comment
  • feat(pkg-func): prefix options with name of pkg-func DavHau/drv-parts#19 (comment)
• A cleaned up package attrset would be nice ahead of
  • [RFC 0092] Computed derivations rfcs#92
• A discussion of the package attrset
  • Redefine outputs in terms of language-level "package", not necessarily store-level derivation (RFC-92, and multi-drv packages, docs) nix#6507
• Reading list
  • [RFC 0035] Default name from pname rfcs#35

[jtojnar]

Do we even need to have a package name for anything other then legacy commands like nix-env? Why not just use the attribute path as the canonical package name?

We do need something to base the store path on (as a convenience when looking at store entries) but keeping the name an opaque string with no codified relation to package/project name will be sufficient for that.