One tree for both runtimes: classic arches to be served by main + reconverse glue, not by resurrecting classic in the branch

[lvkale]

Two hygiene problems on the reconverse-specific-build branch, noted while working at tip 90ef159 (now cdaf6a4):

1. Classic arches no longer build from this branch. ./build charm++ netlrts-darwin-arm8 smp installs reconverse's include/converse.h into the netlrts build dir, so ck-core fails at init.C:1669 with _Cmi_sleepOnIdle / _Cmi_forceSpinOnIdle undeclared. If classic builds from this branch are meant to work, the header install selection needs an arch guard.

2. examples/charm++/hello/1darray/hello.C carries committed debug hackery from 9e48ce9 ("merge reconverse changes with lb"): CkExit() inside the Hello constructor and setInitCallback commented out. The example prints "Hello 0 created" and silently exits 0, which masquerades as a runtime failure to anyone using it as a smoke test. Should be reverted to the mainline version.

(Context: the branch is the reconverse development line, so item 1 may be tolerated deliberately — filing so the state is recorded.)

[lvkale]

Update on item 1: the header clobber had a one-line cause — buildcmake defaulted opt_reconverse=1 ("fake argument for now"), so every build, netlrts included, was configured as a reconverse build. Fixed on the branch along with a conv-static dependency typo. However, with those fixed, configure exposes the real depth: the classic converse library build is commented out in cmake/converse.cmake on this branch (converse is now a custom target wrapping the reconverse external), so classic arches cannot build from this branch without restoring the whole classic library build. Recommendation: treat one-tree-both-runtimes as a property of main + this branch's reconverse glue (where the classic build is intact) rather than resurrecting classic inside the branch.

[lvkale]

Status and reframe:

• Item 2 (polluted hello example): fixed — restored to the mainline version, e498b426a on the branch.
• Item 1 (classic arches unbuildable from the branch): reframed rather than fixed. The shallow causes are repaired on the branch (buildcmake no longer forces RECONVERSE=ON into every build; conv-static dependency righted), but the classic converse library build itself is commented out in cmake/converse.cmake here — restoring it inside the branch would duplicate what main already has intact. The plan this issue now tracks: make one-tree-both-runtimes a property of main + this branch's reconverse glue (reconverse integrated as an arch family, pinned to a reconverse SHA), by landing the branch's changes on main as a curated series. This issue stays open as the tracker for that property until the convergence work supersedes it.