This PR removes uses of `Lean.RBMap` in Lean itself.
Furthermore some massaging of the import graph is done in order to avoid
having `Std.Data.TreeMap.AdditionalOperations` (which is quite
expensive) be the critical path for a large chunk of Lean. In particular
we can build `Lean.Meta.Simp` and `Lean.Meta.Grind` without it thanks to
these changes.
We did previously not conduct this change as `Std.TreeMap` was not
outperforming `Lean.RBMap` yet, however this has changed with the new
code generator.
This feature produced counterintuitive behavior and confused users.
See discussion at #770.
As pointed out by @tydeu, it is not too much work to write `Id.run <|`
before the `do` when we want to use the `do` notation in pure code.
closes#770
After this commit, we have to use an explicit `discard` in code such as
```
def g (x : Nat) : IO Nat := ...
def f (x : Nat) : IO Unit := do
discard <| g x -- type error without the `discard`
IO.println x
```
Motivation: prevent users from making mistakes such as
```
def f (xs : Array Nat) : IO Unit := do
xs.set! 0 1
IO.println xs
```
when they meant to write
```
def f (xs : Array Nat) : IO Unit := do
let xs := xs.set! 0 1
IO.println xs
```
@Kha This one required a bunch of manual fixes. The main issue is that
before we added the string interpolation feature, we created
`MessageData`s using `++` and coercions. For example, given
`(e : Expr)`, we would write
```
let msg : MessageData := "type: " ++ e
```
and rely on the coercions `String -> MessageData` and
`Expr -> MessageData`, and the instance `Append MessageData`.
However, heterogeneous operators "block" the expected type propagation downwards.
This kind of code is obsolete now since we can write a more compact
version using string interpolation
```
let msg := m!"type: {e}"
```
