mirgen: improve scope-only block support

[zerbina]

Summary

Don't emit mnkBlock MIR nodes for nkBlockStmt/nkBlockExpr nodes
(block statements/expression) are not used for control-flow. This
reduces the size of the produced MIR code and means that neither
compiler-generated scope-only blocks nor block statements/
expression not targeted by break statements reach the code
generators.

Most notably, the code generated by the JavaScript code generator
improves, in terms of size.

Details

Detection of blocks-used-for-control-flow is done via testing for the
sfUsed flag: if present, the nkBlockStmt/nkBlockExpr is treated
as used for control-flow, otherwise it's treated as only opening a
scope. This is only a conservative approximation, however, and blocks
where the label is marked as used through other means (or by appearing
in a break within a compiles) will be mis-detected.

So that the detection also works for transf-injected breaks and
blocks, all nkBreakStmt AST in transf / closureiters /
lambdalifting
is now generated via the new newBreakStmt procedure in lowerings,
which marks the provided label as used.

When mirgen.genBlock gets passed an nkBlockExpr/nkBlockStmt node
where the label symbol is not marked as used, it only emits the
mnkScope block necessary for correct variable lifetimes. Since
cnkBlockStmt are only generated for mnkBlock nodes by cgirgen,
this means that the block statements are omitted in the final generated
code too.

Finally, transf.transformBlock is improved, and a necessary fix is
applied:

• instead of introducing a new label with newLabel during inlining,
  the original symbol is copied and adjusted. This ensures that the
  symbol flags (e.g., whether the symbol is used) are kept
• the label symbol is no longer pushed to the breakSyms stack in the
  inlining case. This was/is unnecessary, as inlined AST is already
  transformed, and so none of the nkBreakStmt nodes within are
  processed
• the non-inlining case doesn't use transformSons, but transforms
  the body AST directly. While having no practical effect at the
  moment, this means that the label slot is no longer treated as a
  symbol-in-a-usage-position

No mnkBlock nodes being emitted for scope-only blocks is also visible
in the --expandArc output, so the tests depending on the output are
adjusted.

[saem]

i like the straightforward nature of using the used flag on the label-sym, and the conosolidated break construction cleans up a fair bit of code.

[saem]

/merge

[github-actions]

Merge requested by: @saem

Contents after the first section break of the PR description has been removed and preserved below:

---

Notes for Reviewers

• a follow-up to transf: lower all unlabeled break statements #918