refactor(ipa): single-source the legacy long-running operation list

[julius-jogela]

Proposed changes

This PR moves the hardcoded list of 17 legacy long-running operationIds out of the foascli merge tagger into a single shared file, so the upcoming IPA-132 LRO rules can consume the same list the tagger uses. Tagging behavior is unchanged.

Jira ticket: CLOUDP-438234

Why

The IPA-132 LRO rules will select on the declared 202 response so they also work on the raw per-service specs in MMS, where x-xgen-long-running-operation does not exist (it is only added at the merge step). Those rules need the same legacy exclusions the tagger applies, without maintaining the list twice.

How

• tools/spectral/ipa/rulesets/functions/utils/legacyLongRunningOperations.json — the single source of truth. Lives inside the ruleset package so it ships to npm consumers.
• The Go map in the merge package is now a checked-in file generated from that JSON (go generate ./internal/cli/merge, generator under internal/cli/merge/gen). go:embed cannot cross the tools/cli module boundary and the released foascli binary cannot read repo files at runtime, so generation is the mechanism.
• A new unit test reads the JSON and asserts set-equality with the generated map, so the two cannot drift; it runs in the existing unit-test CI step.

Testing

• go build ./..., go vet, go test -race -cover ./internal... — all pass; the existing tagLongRunningOperations tests are unchanged and green.
• golangci-lint run (v2.10.1, the CI pin): 0 issues.
• go generate verified idempotent; regenerated output is byte-identical to the previous hand-written map.
• Repo-wide prettier check passes (covers the new JSON).

Checklist

• I have signed the MongoDB CLA
• I have added tests that prove my fix is effective or that my feature works

[yelizhenden-mdb]

Q: Does this command run automatically as part of the build, or do we need to run it manually?
(If manual, it would be better if we can add it to somewhere in the contributing.md)

[julius-jogela]

It's manual, go generate doesn't run as part of the build. There's a unit test that compares the generated map against the JSON so the two can't drift. Also added a line to CONTRIBUTING.md for clarity.