Stop shipping HTTPDSL and HTTPJES2

[mgrossmann]

Closes #157

Retires the two modules mvsMF replaced. Build goes from 6 load modules to 4, and the internal archive from 112 objects to 100.

What moved, and why moving matters

[internal] globs src/*.c, so dropping the [[module]] entries alone would have left both clusters compiling into the archive with nothing rooting them. The sources are in tbd/ now — outside that glob — which takes them out of the build rather than merely un-autocalling them.

tbd/src/      httpdsl.c httpdsld.c httpdsle.c httpdsli.c httpdsll.c httpdslp.c
              httpdslx.c httpjes2.c jespr.c jesst.c jestime.c stck2tv.c
tbd/include/  httpds.h jestime.h
tbd/README.md

All via git mv, so history follows. Moved rather than deleted, per the issue — tbd/README.md records what each file was and what the link test showed, because a directory called tbd holding 3,000 lines with no explanation is a puzzle for whoever opens it next.

How the exclusive set was determined

Link test, not reading. Two attempts at static analysis of the call graph gave contradicting answers — the sources use two different function-definition layouts, and CGI modules reach the server through the HTTPX vector rather than by symbol, so a naive graph is both incomplete and misleading. So instead: take the candidates out, rebuild, and let an unresolved symbol speak up, since autocall pulls only referenced members.

All 4 remaining load modules and all 11 test modules link without any of them. The clusters are cleanly separated — httpds_* appears only in the DSL files, clibjes2.h only in the JES ones.

stck2tv.c came along for a different reason: nothing in the tree has ever referenced it, not even the two modules being retired. The same test surfaced it.

The ABI check

The HTTPX vector exports none of these functions, so no external CGI — mvsMF included — can reach them at runtime. Worth recording because the vector's offsets are frozen: had one been exported, this would have broken the published ABI instead of shrinking the build.

Samplib

• httpprm0: the two commented MOD= lines are gone. Also the TZOFFSET example — that keyword was retired in Parmlib TZOFFSET never reaches a module: __tzset() writes the per-task CRT, so CGI timestamps render with the system CVTTZ instead #145 and the sample was the last place still suggesting it. Bundled here deliberately rather than left for a third pass at the same file.
• httpd (STC PROC): HASPCKPT and HASPACE1 removed. Those are the JES2 spool datasets HTTPJES2 read; nothing else in the server opens them, so they would otherwise sit in every installation's PROC pointing at SYS1.HASPACE for nothing.

Docs

Three places went stale the moment the modules stopped shipping:

• docs/configuration.md — listed both as configurable modules with MOD= syntax. Now states that such a line parses but fails to load the program on the first matching request.
• docs/migration.md — said they were "still present in the code". Now names the two concrete migration steps: drop the MOD= lines, drop the two DDs.
• CLAUDE.md — documented the endpoints as usable. Also loses the line calling /jes/status a quick JES2 cross-check, which is mvsMF's job now, and the jesst.c printf-to-stdout note, which described a file that is no longer built.

Verification

make and make test clean under -Wall -Werror: 4 modules, 11 test modules, archive at 100 objects. That is compile-and-link — which is exactly the property at issue here, since the question was whether anything still resolves against these files. No runtime behaviour changed for the four surviving modules.

Not verified on MVS, and it does not need a deploy to be judged: the change removes code from the build rather than altering any code path that runs.