Go back to inlining during typing

[odersky]

This is a tentative PR to move inlining back to typer. This will be needed if we want to address #5381.

Reverts the following commits:

Join containsQuotesOrSplices and containsInlineCalls (reverted from commit 8cf1385)
Make InlineCalls an object (reverted from commit ad26554)
Remove Inlined and InlineProxy from TASTy (reverted from commit 976e095)
Move inlining inside Reify quotes

Split some neg tests as now only the first error is emitted. (reverted from commit 21e65f1)
Erase rhs of erased non inlined val/def in PostTyper (reverted from commit c12e875)
Add isInlineCall to TreeInfo (reverted from commit 954c25b)
Update doc (reverted from commit 0832e02)
Only run InlineCalls if the tree contains an inline call (reverted from commit f28e2e1)
Check if unpickled tree has inline nodes (reverted from commit 927ae4e)
Add a bit of documentation (reverted from commit da0c249)
Fix constant folding during inlining (reverted from commit 32c8798)
Move inline β-reduction after Pickler (reverted from commit c68dc1f)
Normalize call at Inlined node creation (reverted from commit 746fdd7)
Move inline β-reduction after post typer (reverted from commit 432eb0a)
Move inline β-reduction out of typer (reverted from commit 6acaf31)

[Duhemm]

Would it be possible to keep the original call somewhere? This information is important:

• for incremental compilation, because the parameters of an inlined call may introduce dependencies that are not visible in the resulting inlined code
• for the IDE, because the user works with trees before inlining.

[odersky]

A potential problem is that this could increase bytecode size significantly. E.g. looking at the trace calls in the compiler, which surround large bodies of code. If we keep the call that code would be duplicated in the Tasty tree.

Currently, we get the calls for trees that are in an edit buffer, but not for others. So findReferences would only show those references that survive in the inlined code, but not other references that are not inlined.

We should think a bit more how we can reconcile these concerns.

[odersky]

Regarding bytecode size: if we have several nested trace-like calls the blowup could be exponential. You can see this sometimes when you print at phase frontend without -Yshow-no-inlines.

[odersky]

Now included in #5383