[BUG]: RequestSpendView in multisig blocks forever — WithTimeout() exists but is never applied to the channel wait #1671

[adecaro]

Fixes #1671: RequestSpendView.Call() in the multisig and boolpolicy identity services could block forever waiting for a co-signer's answer, since the timeout field/WithTimeout() setter
on multisig.RequestSpendView were dead code (never applied to the wait loop), and boolpolicy.RequestSpendView had no timeout knob at all.

• Extracted the duplicated fan-out/collect pattern into a new generic utils.AnswersCollector[K, T] (token/services/utils/collector.go), which bounds the wait with a per-call timeout
  (defaulting to 30s if unset) and honors context cancellation, returning a descriptive wrapped error either way.
• Rewired multisig/spend.go to actually apply its timeout/WithTimeout() via the new collector, and added the same timeout field + WithTimeout() setter to boolpolicy/spend.go, which
  previously had none.
• Added regression tests for both packages proving Call now returns promptly (instead of hanging) when a co-signer/co-owner never responds, plus a happy-path test, and unit tests for
  the collector itself (all-answers, worker error, timeout, context cancellation, late-send-after-return).

Additional changes

• Enforced a new coding rule: never use fmt.Errorf/fmt to construct or wrap errors — always use github.com/hyperledger-labs/fabric-smart-client/pkg/utils/errors (New, Errorf, Wrap,
  Wrapf, WithMessage, WithMessagef, Join, etc.). Documented this in AGENTS.md and fixed the one violation found in docs/upgradability.md, plus rewrote collector.go's two error sites to
  use errors.Join/errors.Wrapf instead of fmt.Errorf.
• Added a "Documentation Updates" workflow rule to AGENTS.md requiring docs/ to be updated (when affected) before any task is reported as complete.