Fix agent not enforcing MFA device lock after in-band MFA.

[cthach]

Part 2 of bug fix for https://github.com/gravitational/core/issues/146.

After successful in-band MFA during SSH session establishment, the Agent will append the MFA device used to satisfy the MFA challenge as a lock target to decisionpb.SSHAccessPermit.LockTargets.

This works because lock enforcement happens further downstream in AcquireSessionContext and NewServerContext; the former being lock enforcement for new connections and latter for established connections.

Alternatives considered

1. Pass the MFADevice through ssh.Permissions.Extensions or ssh.Permissions.ExtraData. Rejected because the SSH access permit is authoritative on lock targets and side stepping that would be confusing and error prone.
2. Create a new field MFADevice on the IdentityContext. Rejected because it was has broader implications and because it conflicts with IdentityContext.UnmappedIdentity.MFAVerified.

Future work

1. Add missing test coverage for locks to TestSSHOnMultipleNodes. Current coverage added in this PR should suffice, but adding integration coverage will give us confidence the whole system is working E2E, especially the trusted cluster scenario. It should be a matter of creating a lock, waiting for the lock to propagate, attempting to connect and asserting on the error.

Manual Test Plan

Test Environment

Teleport instance built with make full-ent running on my Mac. My clients are built from Teleport v18.

Test Cases

• SSH to node with per-session MFA should succeed
• SSH to node with per-session MFA and device lock should reject the new session
• SSH to node with per-session MFA, no device lock initially, and enable device lock after session established, should terminate the established session

[cthach]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Already looking forward to the next diff.

Reviewed commit: c82aa74d96

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[cthach]

TODO: Migrate this PR to our new repo. Will open it for review then.