[Feature]: Transports — LocalTransport + SSHTransport

[aldbr]

User Story

As the BatchBackend (and any composed scheduler),
I want Transport implementations for local execution and SSH,
So that reaching a resource is an orthogonal, swappable axis (IC-ADR-001 §3) validated against the
real sshd stack.

Feature Description

• Transport protocol finalised: run(argv) -> CommandResult,
  put(local, remote), get(remote, local); timeouts; typed errors (auth vs connectivity vs
  command failure); async context-manager lifecycle (connections are held state — DIRAC's
  shutdown() closes ssh gateways today).
• LocalTransport: asyncio subprocess exec + local file ops.
• SSHTransport: async SSH library; key auth; pinned known-hosts; connection reuse across bulk calls; optional
  jump/gateway host (parity with DIRAC's SSHTunnel).
• Design consequence to state in docs: unlike DIRAC (which ships a stdlib-only python driver
  to the remote host and execs it), interCEde builds commands locally and sends only argv —
  no python required on the remote host (ADR-REVIEW.md comment 9).
• Integration: transport-level tests (run/put/get, quoting, timeouts) against the ssh-slurm
  stack's sshd; local variant against the container itself.

Definition of Done

• LocalTransport + SSHTransport pass a shared transport conformance suite (unit + against
  the sshd stack)
• Quoting/escaping torture tests (spaces, quotes, globs in argv and paths)
• Jump-host path covered (config + at least a unit-level test)
• Registered via intercede.transports entry points
• Credential supply follows IC-ADR-003 (#) for SSH keys

Alternatives Considered

• Fabric/Paramiko (DIRAC's choice) — sync; would force asyncio.to_thread around every call.
• Keeping DIRAC's ship-a-python-driver model — requires python2/3 on every remote host and
  stdlib-only drivers; dropped by the ADR (commands over the wire instead).

Related Issues

Additional Context

SSHMultiHostTransport (DIRAC SSHBatch: N hosts + per-host slot spreading) is deferred —
placement policy vs transport question flagged in ADR-REVIEW.md comment 7.