e2e: an empty TIA selection concludes "E2E Tests" green with no suite run

hack/select-e2e.sh emits an empty selection for a changed path it does not recognise outside packages/. Both e2e lanes read an empty selection as "nothing to run": the Chainsaw step is skipped and the required E2E Tests status is posted success with zero suites executed. This affects pull-requests.yaml and e2e-fork.yaml equally.

Skipping is the intended behaviour for genuinely inert paths — a PR touching only .gitignore should not burn a full e2e run. The defect is that the silent-ignore branch currently swallows paths that are not inert. full_suite_pattern covers hack/[^/]+\.sh$ and hack/[^/]+\.bats$ but not hack/*.mk, so a change to hack/common-envs.mk — which supplies the tag, push and output flags of every image in the tree — selects nothing and skips e2e entirely. .github/workflows/e2e-fork.yaml is in the same position: pull-requests.yaml is in the pattern, the fork lane's own workflow is not.

A blanket "empty selection means full suite" fallback is the wrong fix, because it escalates exactly the inert paths the design deliberately skips. What this needs is an audit of the currently-ignored paths and an explicit classification:

affects every build (hack/*.mk, shared Makefile fragments) → full suite
affects an e2e lane itself (.github/workflows/e2e-fork.yaml) → full suite
no runtime impact (.gitignore, editor config, renovate.json, *.md outside docs) → keep skipping, but by an explicit rule

Turning the third group into an explicit skip-list rather than a silent default is the part that matters: anything matching neither list would then escalate instead of quietly selecting nothing, so the next path someone adds fails safe rather than fails open.

This is pre-existing behaviour and not a regression from #3262 — it was found while reviewing that PR, at #3262 (comment).

docs(ci): correct the select-e2e.sh escalation claim in e2e-fork (#3262)

One addition from review work on the fork e2e lane: there are now two consumers of an empty TIA selection, not one. The in-tree e2e job and the fork lane's resolve/e2e pair both read an empty selection as "skip Chainsaw" and still conclude green, so the fix should cover both call sites. The escalate-unclassified change (#3511) closes the in-tree half and, once the fork lane lands (#3262), applies to it as well, since both feed the same hack/select-e2e.sh.

fix(ci): classify every TIA path and escalate the unclassified

fix(ci): escalate unclassified paths in TIA instead of skipping e2e (#…