AssocList.replace: Even less allocations

[nomeata]

@crusso tempted me to look at #535, and I found some more improvements, by avoiding to reconstruct the (hd_k, hd_v) tuple.

Before:

{heap_diff = 16_080_000; instructions = 92_841_876}

After:

{heap_diff = 8_088_000; instructions = 69_362_716}

[nomeata]

I also found the problem why the old code was allocating. The same effect as #535 could be achieved with

diff --git a/test.mo b/test.mo
index 2c52719..65bb9ac 100644
--- a/test.mo
+++ b/test.mo
@@ -14,8 +14,8 @@ actor {
       switch (al) {
         case (null) {
           switch value {
-            case (null) { (null, null) };
-            case (?value) { (?((key, value), null), null) }
+            case (null) { return (null, null) };
+            case (?value) { return (?((key, value), null), null) }
           }
         };
         case (?((hd_k, hd_v), tl)) {
@@ -23,12 +23,12 @@ actor {
             // if value is null, remove the key; otherwise, replace key's old value
             // return old value
             switch value {
-              case (null) { (tl, ?hd_v) };
-              case (?value) { (?((hd_k, value), tl), ?hd_v) }
+              case (null) { return (tl, ?hd_v) };
+              case (?value) { return (?((hd_k, value), tl), ?hd_v) }
             }
           } else {
             let (tl2, old_v) = rec(tl);
-            (?((hd_k, hd_v), tl2), old_v)
+            return (?((hd_k, hd_v), tl2), old_v)
           }
         }
       }

Why? Because we support unboxed tuple returns from functions, but not from blocks! So a switch block cannot return multiple values.

This should hopefully work out of the box once we can use the multi-value extension (dfinity/motoko#1459)

The relevant backend code is

https://github.com/dfinity/motoko/blob/409363d6bef896a3ebbd041c432e3233a12160b0/src/codegen/compile.ml#L9781-L9796

There is probably a reason I am not using a local and not a branch instruction to exit the block, which I do not recall at the moment.

[nomeata]

In dfinity/motoko#3566 I improved the situation for if; arguably switch should behave the same.

[nomeata]

This is part of #539