Fix crash in device_map='auto' when CPU is missing from inferred max_memory

[pallavisharma6802]

What does this PR do?

Fixes a crash when using device_map="auto" in environments where "cpu" is missing from the inferred max_memory dictionary.

Bug

When max_memory is not explicitly provided, accelerate.get_max_memory() may return a dictionary without a "cpu" key (depending on the backend/environment).
_get_device_map then unconditionally scales "cpu" memory, resulting in a KeyError.

Fix

• Guard access to "cpu" in inferred_max_memory before applying the scaling factor.
• Preserve existing behavior when "cpu" is present.

Tests

• Added tests/utils/test_device_map_cpu_guard.py to ensure device_map="auto" does not crash when "cpu" is absent from max_memory.

Fixes #42994

Before submitting

• Did you read the contributor guideline?
• Did you write any new necessary tests?

Who can review?

@Cyrilvallez

[pallavisharma6802]

Thanks for the report! This PR adds a defensive fix for the device_map="auto" crash and includes a regression test.
Happy to make any adjustments if needed.