[k8s] fix managed job issue on k8s

[nkwangleiGIT]

1. initialize a new dict if the destination is None, and add the source key/value to it
2. use the cpu/memory to k8s resources limits, or the pod will fail to create if there is LimitRange configured in the namespace, as the default cpu/memory maybe smaller than the requests.

Tested (run the relevant ones):

• Code formatting: bash format.sh
• Any manual or new tests for this PR (please specify below)
• All smoke tests: pytest tests/test_smoke.py
• Relevant individual smoke tests: pytest tests/test_smoke.py::test_fill_in_the_name
• Backward compatibility tests: conda deactivate; bash -i tests/backward_compatibility_tests.sh

[romilbhardwaj]

Thanks @nkwangleiGIT!

[romilbhardwaj]

thanks for adding this!

[romilbhardwaj]

We do not set limits because we want SkyPilot pods to use any idle CPU cycles instead of being throttled to a specific limit. https://home.robusta.dev/blog/stop-using-cpu-limits

Suggested change

	cpu: {{cpus}}
	memory: {{memory}}G

[nkwangleiGIT]

ok, let me remove it, maybe we should add this notice to docs somewhere.

[romilbhardwaj]

Thanks @nkwangleiGIT!