Harden the release preflight, and bring the docs back in line with the tests

[wu-sheng]

Follow-up to #194 and #195, before 5.0.0 is cut.

Release scripts

The skywalking-swck release script died on git commit with nothing staged — non-zero under set -e, no message. I cross-checked the same class of failure here by running each construct rather than reading it. Three were real:

Construct	Behaviour	Fix
VERSION=$(grep '^version: ' Chart.yaml | awk …)	grep misses → pipefail → set -e kills the assignment silently, and the die below it is unreachable	awk, which exits 0 either way
read -r reply in remove_previous	exits 1 at EOF → run dies with no output, after the irreversible svn promotion	both reads now report why
trap … RETURN in upload_to_svn	does not fire when set -e kills the shell mid-function → leaks a temp dir holding the signed artifacts	EXIT

Worth noting the first one: false && die "…" as a bare statement does survive set -e, so the tag check was fine — the failure was specific to the assignment. I verified both.

Preflight checks the tools are usable, not just present

Each of these otherwise fails late, at the worst possible point:

• no gpg secret key → fails after the entire build and package
• unusable svn credentials → fails after the tag is pushed
• unauthenticated gh → fails after the svn mv into dist/release, which cannot be undone

Missing tools are now reported in a single message rather than one per run.

The helm gate is 3.8+, not 3.x: 3.8 is where helm push to an oci:// registry landed, and the chart ships only as an OCI artifact. Helm 4 is accepted — Chart.yaml is apiVersion: v2. I confirmed that by running dep up (including the OCI pull of banyandb-helm 0.7.0), package, lint and template through helm 4.1.1; all six storage permutations render, rc=0.

Ordering matters in one spot: the check for a half-finished svn upload runs only after reachability is established, because an auth failure and a not-yet-uploaded version produce the same non-zero exit.

Docs

• contributing/e2e-tests.md described five cells including two OAP 10.4 ones. There are four, all OAP 11 — and the fourth, the config-override cell, was undocumented. Rewritten, with an explicit note that the 10.4 combination is documented but not verified.
• Removed two orphaned expectation files (horizon-admin-readonly.yml, horizon-templates-readonly.yml) that no cell referenced.
• changes/changes.md predated Configure Horizon and OAP by environment variable, mount a file only on request #195: it described the pre-Configure Horizon and OAP by environment variable, mount a file only on request #195 ui.config model, omitted oap.extraEnv / oap.envFromSecret, and — most significantly — never mentioned that the SWCK charts were removed, which is the most disruptive change in 5.0.0.
• Documented the new preflight checks and the reasoning about when they fail.

Verified: all internal links resolve, menu.yml matches the tree, every e2e YAML parses, helm lint passes.