internal/gocore: have single entry point for makeMemRoot (refactor)

I was reading through the code and found that getting all references
(places where roots are made) was made a little more difficult by having
a separate `p.makeMemRoot` versus `makeMemRoot`.

I can understand why it was done: to make `gocore.Core(...)` function
rely more on `core.Process` and make it clearer what the dependencies
are. But the `nRoots` passing is awkward.

Change-Id: I3d8f6f8d9c46a09740a4d28fbe742e0af27b806a
Reviewed-on: https://go-review.googlesource.com/c/debug/+/748440
Reviewed-by: Keith Randall <khr@golang.org>
Auto-Submit: Nicolas Hillegeer <aktau@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>