2026 ithread repairs

[joesuf4]

This patch enables

1/ true ithread multiplexing for HTTP/2

2/ real resilience under h2load across thousands of concurrent theads/connections.

3/ full support for segfault-free live interpreter destruction under such loads.

4/ fixes the dumbness around pulling an ithread interpreter out of the pool just to run the request cleanup hook, which fires on every request: even on static requests that have nothing to do with mod_perl.

It also disables support for PerlInterpMaxRequests, which was always a dumb idea.

With this patch (and the corresponding mod_http2 patch), I can sustain 1200 req/s under a 10K concurrency load (100 connections x 100 HTTP/2 multiplexed channels) for over a minute without failure. Without this patch set, you can't come anywhere close on a 4 core machine with 32 GB RAM.

[joesuf4]

I've beaten the living shit out of this PR, and it fleshed out a pair of serious bugs in mod_http2 as a result.

It now only cores under serious pressure on interpreter destruction, which is tunable to taste. I've always recommended postponing interpreter destruction to global httpd process destruction. This is much more robust than that, but it's the best I can do at this juncture.