Fix EFA field naming

[FarhanTejani]

What's changing and why?

Closes #306

The v1.1 schema and model defined EFA fields as efa_interfaces/efa_interfaces_limit, but the jinja template used efa/efa_limit. This caused:

• efa in config YAML to be rejected by validation ("Extra inputs are not permitted")
• efa_interfaces in config YAML to pass validation but not render in the template

Standardized on efa/efa_limit to match the naming pattern of other resource fields (accelerators, vcpu, memory) and the Kubernetes resource key (vpc.amazonaws.com/efa).

Before/After UX

Before:

Specifying efa in config.yaml fails validation:

efa: 4
# Config validation errors:
#   efa: Extra inputs are not permitted

Specifying efa_interfaces passes validation but is silently ignored in the rendered YAML.

After:

EFA renders correctly in the K8s YAML:

efa: 4
efa_limit: 4

resources:
  requests:
    vpc.amazonaws.com/efa: 4
  limits:
    vpc.amazonaws.com/efa: 4

How was this change tested?

• All unit tests pass
• Added jinja template rendering tests to catch schema / template field name mismatches

Are unit tests added?

Yes. Added TestJinjaTemplateRendering with two tests that render the jinja template directly and verify all resource fields (accelerators, vcpu, memory, efa, accelerator partitions, etc.) appear correctly in the output. This catches the class of bug where schema field names drift from template variable names.

Are integration tests added?

N/A

Reviewer Guidelines

‼️ Merge Requirements: PRs with failing integration tests cannot be merged without justification.

One of the following must be true:

• All automated PR checks pass
• Failed tests include local run results/screenshots proving they work
• Changes are documentation-only

[mollyheamazon]

Should add this:

   if self.accelerators or self.accelerators_limit:
        raise ValueError("instance_type is required when specifying accelerator resources")

This should be part of PR #393 since it is fixing issue #317

[mollyheamazon]

The efa field naming change looks good.
The other fix in the PR is that it removes the node_count vs resources mutual exclusivity check. This was discussed in thread link — please confirm with the stakeholders to align with this approach given the quota implications, or if you'd prefer we introduce --num-replicas instead.

[FarhanTejani]

Scoped this PR down to just the EFA field rename (#306) and template rendering tests. I'll open a separate PR for the node_count work after syncing with the stakeholders.