Ship static/index.html as a UFS webroot image, uploaded with IND$FILE

[mgrossmann]

static/index.html is in the repository and in no release artifact. The
4.0.0 package installs a server whose DOCROOT=/www has nothing behind it.

Static content lives on a UFS disk, and a UFS disk cannot travel the way the
rest of the package does:

• SMP cannot carry it. There is no element type for a DSORG=PS, RECFM=U, BLKSIZE=4096 image (++MOD/MAC/SRC/MACUPD/SRCUPD/ZAP is the whole set), and
  it is site content an APPLY must never touch.
• XMIT cannot carry it either. xmit370 create packs a directory into a
  PDS with --recfm fb|f, LRECL 80. Beyond the tooling: TSO RECEIVE allocates
  its own target and refuses to merge into an existing dataset, so it is a
  first-install-only transport for something that gets updated.

So the disk ships as a file in the dist archive and is uploaded with
IND$FILE from the operator's 3270 emulator — the one path that needs
neither mvsMF nor ftpd on the target.

Verified before writing any of this

• IND$FILE is present on the reference system (SYS2.CMDLIB).
• Binary is the default (no ASCII/CRLF), and RECFM then defaults to U
  (ind_file370 c/indparse.c:249-256).
• Blocking is exact: RECFM=U goes through __fputc, which flushes at BLKSIZE
  (libc370 src/clib/@@fputc.c:24), so BLKSIZE(4096) writes 4096-byte
  blocks. A 1 MB image is 256 whole blocks, no remainder.
• A pre-allocated dataset is the safe path: IND$FILE allocates an existing
  dataset DISP=SHR and takes its DCB, and only falls back to NEW CATALOG when
  it does not exist (ind_file370 c/indmain.c:190-200, :273). That makes the
  upload independent of which IND$FILE build the system carries.
• Wrong blocking is caught at mount: ufsd checks 1 << blksize_shift against
  the dataset BLKSIZE and answers UFSD062E SUPERBLOCK VALIDATION FAILED
  (ufsd src/ufsd#sbl.c:69).
• DISP=SHR is the trap: IND$FILE will rewrite a dataset UFSD has mounted.
  UNMOUNT PATH=/www has to be a numbered step, not a footnote.
• Encoding needs no decision and no conversion. http_send_file() translates
  UFS files with the hard-coded IBM-1047 table, independent of CODEPAGE=
  (src/httpfile.c:70-71), which is exactly what ufsd-utils cp writes.
  Measured end-to-end: ufsd-utils cp -> stored bytes -> ibm1047_etoa gives
  back static/index.html byte for byte, UTF-8 sequences included. Over all
  256 byte values only 0x85 and 0xF7 fail to round-trip.
• httpget.c:68 appends index.html for a directory request, so / serves
  the shipped page.

What to do

• make webroot builds dist/httpd-webroot.img from static/ with
  ufsd-utils (pinned release, fetched into .mbt/tools/ when not on
  PATH — CI runs the shared mbt workflow and cannot install it otherwise).
  1 MB = 256 blocks, 62 inodes; explicit --owner/--group so the
  artifact does not carry the build machine's userid.
• package and dist depend on it; [distribution] extra puts it in the
  zip/tar.gz. Publish the SHA256 rather than claiming reproducibility --
  the image carries creation timestamps.
• samplib/httpwebr: allocation job for HTTPD.@VRM@.WEBROOT.UFS
  (SPACE=(4096,256), DCB=(DSORG=PS,RECFM=U,BLKSIZE=4096), primary
  extent only, mirroring ufsd's UFSFMT ALLOC step).
• docs/installation.md: a numbered section for the webroot disk --
  allocate, upload with IND$FILE, mount at /www, verify, and the
  unmount-before-re-upload rule.
• smp-todo.md O3: replace U1-U6 with the decided path.

Not in this change

CODEPAGE=IBM1047 in the shipped HTTPPRM0 was considered and rejected on
measurement. cc370 compiles string literals in CP037 (common/src/mvs370.c
mvs_a2e_tab), and CODEPAGE= governs exactly those. Under IBM1047 a [
compiled as 0xBA leaves as 0xDD, which would break every mvsMF array
response -- the primary route in the shipped configuration -- plus httpd's own
login form and error pages. Live check on the reference system: [ currently
arrives as 0x5B under the CP037 default. Static UFS files are unaffected
either way because they never consult the setting.

[mgrossmann]

Merged as PR #253 into release-4.0.0-smp-distribution (fa65b47) — the branch is not the default one, so the Fixes line did not fire on its own.

The archive now carries httpd-webroot.img; samplib/httpwebr allocates the dataset, and docs/installation.md §9 is the operator's path: allocate, IND$FILE, mount, verify, and unmount before replacing. The image is built under build/ rather than dist/, so it travels in the archive instead of appearing as a version-less Release asset beside it.

Encoding needed no decision in the end: UFS files are translated with the hard-coded IBM-1047 table (src/httpfile.c:70) independent of CODEPAGE=, which is what ufsd-utils writes — measured byte-exact. The keyword's undocumented scope became #254.

Not verified on MVS: the 3270 transfer itself, and the Linux ufsd-utils fetch that release CI will do. make prerelease exercises the second one.