[Improve] Frame provider comment follow-ups as untrusted content

[mrubens]

Description

Follow-up to #570, which added shared prompt-hardening helpers (buildMentionRequestBlock, buildUntrustedContentPolicy) and applied them to the GitHub-facing prompts. The non-GitHub provider comment handlers still interpolated the triggering comment into follow-up task messages with plain > quoting and no untrusted-content policy.

The GitLab, Bitbucket, Azure DevOps, and Gitea comment handlers now compose their existing-task follow-up messages the same way as the GitHub issue-mention handler:

• The triggering comment (authored by a commenter who passed the linked-account gate, so it is the request to act on) lands in an escaped <mention_request> block instead of > quoting.
• The PR/MR title is escaped with escapeTaskContextText, matching the GitHub composition.
• The shared untrusted-content policy is appended to each built follow-up message.
• The now-unused local formatQuotedText helpers are removed.

Test plan

• Handler tests now mock the @roomote/cloud-agents/server barrel with distinctive marker fakes and assert the follow-up message routes text through the right builders, mirroring the GitHub handler test update in [Improve] Frame third-party text in agent prompts as untrusted content #570.
• The Bitbucket suite gains its first coverage of the reusable-task follow-up path (steer into an active owner task).
• Ran the four affected apps/api handler test directories (21 files, 146 tests), pnpm lint, pnpm check-types, and pnpm knip.

[roomote-roomote]

No code issues found. See task

_{Reviewed 6ac7c0f}