[Bug]: Fleet and run GPU names are combined case-sensitively, dropping all backend offers #4214

Steps to reproduce

1. Apply a fleet whose resources.gpu name is written in lowercase:

type: fleet
name: my-fleet
nodes: 0..
resources:
  gpu: h100:1..

2. Apply a run configuration requesting the same GPU with a different case:

type: task
name: my-task
commands:
  - true
resources:
  gpu: H100:1

Actual behaviour

The run plan shows no backend offers. Only offers from instances that already exist in the fleet are listed, so a run that has no reusable instance to land on gets No matching instance offers available. Writing h100 instead of H100 in step 2 brings the backend offers back.

Backend offers are collected per candidate fleet, and the fleet's requirements are intersected with the run's by _combine_gpu():

The GPU name intersection is a plain case-sensitive in test:

So ["h100"] ∩ ["H100"] yields [], and an empty (non-None) name list matches no GPU at all, which drops every backend offer the fleet could contribute. Offers from existing instances survive because that path matches GPU names case-insensitively (gpu_matches_gpu_spec(), gpuhunt.matches()), which is also why dstack offer --gpu H100:1 still returns the offers - it ignores fleets.

Expected behaviour

GPU names are compared case-insensitively when combining fleet and run requirements, as they are everywhere else in offer matching. dstack offer and dstack apply should agree.

dstack version

master (db06abf)

Additional information

_intersect_lists_optional() is also used for regions, availability_zones, and instance_types when combining fleet and run profiles, and all three are lowercased before offers are filtered in _filter_offers(), so they have the same latent problem. backends is an enum and is unaffected.

Minimal check without a backend:

from dstack._internal.core.models.resources import GPUSpec
from dstack._internal.server.services.requirements.combine import _combine_gpu

fleet = GPUSpec(name=["h100"], count="1..")
print(_combine_gpu(fleet, GPUSpec(name=["h100"], count="1")).name)  # ['h100']
print(_combine_gpu(fleet, GPUSpec(name=["H100"], count="1")).name)  # []