build: fix stale soname symlink after a library version bump

[grandixximo]

When a library soname is bumped (here liblinuxcncini.so.0 to .so.1), an incremental tree keeps the old .so.0 on disk, and the dev symlink ../lib/liblinuxcncini.so gets repointed at the stale lib: the generic ../lib/%.so pattern rules in src/Makefile match both %.so.0 and %.so.1, and the .so.0 rule is listed first so it wins. Everything then links against the old library and fails with undefined symbols (IniFile::mapLinearUnits), breaking the python linuxcnc module, linuxcnc_check_ini and startup.

Clean builds are immune, so CI never sees it; only people building incrementally across the bump are hit. make clean does not help either, because .so.0 is no longer in $(TARGETS) so rm -f $(TARGETS) leaves it behind.

Fixes:

• the liblinuxcncini.so.1 rule removes any leftover .so.[0-9]* sibling.
• an explicit liblinuxcncini.so to .so.1 symlink rule, which overrides the ambiguous pattern pair so the symlink repoints to the current soname in a single build pass (the rm alone only self-heals on a second make).
• genclean also removes ../lib/*.so.[0-9]* so stale versioned libraries no longer survive make clean.

@BsAtHome flagging you as reviewer. AndyPugh, myself, and Peter Wallace from Mesa all fell into this same trap; I walked Peter through it on Discord. A general fix for everyone seems nicer than the git clean -dxf we keep recommending, and it avoids that step entirely. Any reason not to do it?

[BsAtHome]

Should there also be a matching fixup in 2.9 to ensure .so.0?

[grandixximo]

liblinuxcncini is the only lib that differs across branches (.so.0 in 2.9, .so.1 in 2.10), so the only one a shared RIP tree holds in both versions. 2.9 => 2.10 is the breaking direction this PR fixes; the reverse is safe since 2.9 only has the %.so: %.so.0 rule and rebuilds .so.0 on checkout, so a stray .so.1 is never picked up. No 2.9 correctness fix needed.

A fully general fix (auto-handling any future bump) is not really feasible without routing every lib through one build macro, since sonames are written inline per-Submakefile. The genclean glob here is the general part (covers all libs); the explicit symlink + sibling-rm is just the cheap per-bump recipe. I backported the genclean glob to 2.9 for tidiness: #4188.

[BsAtHome]

Check.