Spec for `${VAR}` interpolation rules inside `connections:`

[rafafrdz]

Motivation

Depends on #1 and #2.

When connection fields accept `${VAR}` references, we need to lock down the resolution rules so every implementation (parser, server, UI preview) behaves identically.

Questions to answer in the spec

1. Resolution order: pipeline variables → environment variables → error? Or env only, or vars only?
2. Timing: are variables resolved at parse time (before reaching the backend) or at execution time?
3. Missing variables: hard error vs. empty string fallback?
4. Escaping: how does a user include a literal `${foo}` in a connection field?
5. Scoping: can a connection reference another connection's resolved value? (probably no.)
6. Security: are resolved values logged/echoed anywhere? Fields marked `sensitive: true` should never appear in logs or the `ExplainPipeline` output.

Deliverables

• A short section in the spec document describing each rule with examples.
• Test fixtures in the spec repo (`spec/v3.0/fixtures/connections/`) covering happy paths and edge cases so implementations can share them.

Related

• Interpolation semantics should match what `teckel-parser` already does for YAML variables, to avoid a second parser.