fix(cli): reject off-tier template cpu/memory pairs before the API

[robertohluna]

Follow-up to #94 (already merged), addressing coordinator feedback on the templates create resource flags.

Problem

The backend only accepts published vCPU/memory pairs:

size	vCPU	memory
xs	1	2048 MiB
small	2	4096 MiB
medium	4	8192 MiB
large	8	16384 MiB
xl	16	32768 MiB

Disk is a free-form floor. After #94, templates create forwarded any --cpu/--memory combination, so an off-tier pair like --cpu 4 --memory 4gb (4 vCPU / 4 GB) was sent and rejected by the platform with a shape-mismatch 500 — after the Dockerfile upload.

Fix

templates create now resolves resources locally (resolveTemplateResources) and refuses off-tier pairs before any API call:

• A --cpu/--memory pair must match a published shape; a missing member is filled from --size (or small).
• Off-tier pair → fail fast with an actionable message naming the nearest size, e.g.:
  cpu/memory 4/4096 isn't a supported pair; nearest is medium (4 vCPU / 8192 MiB); pass --size medium
• A pair that conflicts with an explicit --size is rejected (match large, not --size small).
• --disk stays free-form; --size <name> alone is always safe; with no cpu/memory pins the fields are omitted so the server defaults.

Verification

• npm run build (tsc): clean.
• npm test: 431 pass (67 files), incl. a new off-tier rejection test; the prior test updated to a valid medium pair (4 vCPU / 8 GB).
• Live (built CLI): off-tier 4/4gb and 3/6gb fail locally with nearest-size hints; 8/16gb --size small reports the conflict; valid 2/4gb and --size large pass validation and proceed to the API. All emit clean JSON.

No version bump.