feat(server): remove the embedded viewer

[NivGreenstein]

Closes MAPCO-11482. Docs half: MapColonies/shigola-docs#2

Shigola serves tiles, not a map. The viewer route, the embedded UI, the build tags that excluded it,
and the viewer half of the observability interface all go.

Sequenced first among the surface removals, as the ticket asked: the viewer consumed both the
capabilities endpoints and the native /maps/... routes, so removing those first would have left it
broken in passing rather than removed on purpose.

What went

Routes	setupViewer and both its build-tagged implementations
Assets	ui/ — the Vue app, its go:embed wrapper, and the go:generate that shelled out to npm
Build tags	noViewer, and the generated files registering it
Observability	ViewerObserver, InstrumentViewerHandler, and both implementers
Metrics	the whole shigola_viewer_* family
CI	the test job's UI step, the release workflow's build_ui job, its ui artifact, and the ui input on shigola-setup-env

The only go:generate left in the tree is the build-tag one, which needs nothing but Go. Nothing
in the build needs node any more.

Three things reviewers should look at

1. A near-miss that PR CI could not have caught. Dropping the ui: true input emptied five
with: blocks in on_release_publish.yml. A bare with: parses as YAML null, which GitHub's
workflow schema rejects — invalidating the entire release workflow. on_pr_push.yml never
exercises that file, so CI would have stayed green and the next release would simply have failed to
parse. Those with: lines are removed rather than left empty; both workflows and the composite
action are verified to parse.

2. The CI scope is deliberately wider than the ticket. The AC names only "the CI step … in the
test job". But the release workflow had a whole build_ui job feeding seven downstream jobs through
a ui artifact. Removing only the test step would have left every release build waiting on an
artifact no job produces, so both halves are gone.

3. The docs do not announce the removal. They simply stop mentioning a viewer. There is no
stable release to upgrade from, so there is nobody a migration note would address, and documenting
an absence only raises a question the reader did not arrive with. This deliberately does not satisfy
the ticket's "noted as a breaking dashboard change" wording — see the note on the ticket.

4. Four generated files here are not about the viewer. AC 4 requires the build-tag files be
"regenerated rather than hand-edited". Regenerating revealed that the committed set was stale:
noGCSCache and noHanaProvider exist in the tree (atlas/cache_gcs.go, atlas/provider_hana.go)
but had never been registered. So no_gcs_cache / no_hana_provider pairs appear here. Pruning them
back by hand would be precisely the hand-editing the AC and CLAUDE.md forbid. shigola version now
reports them, and README's build-tag list — which had never documented either — is corrected to match.

On ADR-0003

AC 7 asks that ADR-0003, which placed the viewer at its own path, be amended. There is no ADR
register. ADR-0001 through ADR-0009 are referenced only from code comments; nothing defines
them in either repo, and a Confluence search found nothing. What is amended here is therefore the
four in-code comments citing ADR-0003. If the register lives somewhere I could not see, it still
needs the real amendment — please say where.

Verification

• Test proven red before green. server/ogc_mount_test.go previously asserted the /viewer
  → /viewer/ redirect; it now asserts 404 on /viewer, /viewer/ and /viewer/index.html
  through a real server.NewRouter. All three failed with 301 before the removal.
• cmd/shigola and cmd/shigola_lambda build; cmd/shigola also builds with CGO_ENABLED=0.
• go build -tags noViewer still succeeds — the tag is simply inert now.
• Full suite green in both CGO modes, gofmt -s clean, all four workflow/action YAML files parse.
• The OGC CITE conformance job runs on this PR (it touches server/server.go and server/ogc/**),
  independently checking the OGC surface still conforms.

Pre-existing failures, not from this change

cmd/shigola/cmd/cache TestValidateTileInGrid/a_zoom_beyond_the_scheme's_matrices and server/ogc
TestTile/a_matrix_the_scheme_does_not_have fail on master today — confirmed by stashing this
branch and re-running. Separately, cache/file TestExpiration failed once under full-suite load
and passes 3/3 in isolation on both master and this branch: a timing flake. All three want their
own ticket.

Deliberately left

CHANGELOG.md's viewer entries are upstream history and are untouched. LICENSE.md's attribution
keeps every word except naming the viewer among what this codebase contains — it no longer does.

🤖 Generated with Claude Code

[coveralls]

Coverage Report for CI Build 0

Coverage increased (+0.06%) to 47.744%

Details

• Coverage increased (+0.06%) from the base build.
• Patch coverage: 6 uncovered changes across 2 files (3 of 9 lines covered, 33.33%).
• 1 coverage regression across 1 file.

Uncovered Changes

File	Changed	Covered	%
internal/build/not_no_hana_provider.generated.go	4	0	0.0%
internal/build/not_no_gcs_cache.generated.go	2	0	0.0%
Total (3 files)	9	3	33.33%

Coverage Regressions

1 previously-covered line in 1 file lost coverage.

File	Lines Losing Coverage	Coverage
internal/observer/null.go	1	0.0%

---

Coverage Stats

Relevant Lines:	20099
Covered Lines:	9596
Line Coverage:	47.74%
Coverage Strength:	206.57 hits per line

---

💛 - Coveralls