Read ACCELERATE_BYPASS_DEVICE_MAP case-insensitively, like every other boolean flag

[Sreekant13]

What does this PR do?

Reads ACCELERATE_BYPASS_DEVICE_MAP case-insensitively, the way every other boolean environment flag in the library is read.

All three sites compared the raw value, so ACCELERATE_BYPASS_DEVICE_MAP=True silently did nothing:

# accelerator.py:1475, 1816, 1890 (before)
os.environ.get("ACCELERATE_BYPASS_DEVICE_MAP", "false") != "true"

It is the only one of eleven such flags read this way. ACCELERATE_USE_FSDP, ACCELERATE_USE_DEEPSPEED, ACCELERATE_USE_MEGATRON_LM, ACCELERATE_USE_SAGEMAKER, ACCELERATE_USE_PARALLELISM_CONFIG, ACCELERATE_ALLOW_CP_STANDALONE, ACCELERATE_DEBUG_MODE, ACCELERATE_DEEPSPEED_ZERO3_SAVE_16BIT_MODEL, FSDP_OFFLOAD_PARAMS and PARALLELISM_CONFIG_SP_SEQ_LENGTH_IS_VARIABLE all .lower() first.

value    BYPASS_DEVICE_MAP   USE_FSDP (sibling, same file)
'true'   True                True
'True'   False               True
'TRUE'   False               True

That inconsistency is what makes it a trap rather than merely strict. True is the spelling a Python user reaches for, and it works for the flag two hundred lines up. When it is ignored the user still hits the device_map='auto' ValueError, whose text does not mention the variable, so nothing points at the value as the cause; the DDP device_ids branch at 1890 takes the non-bypass path for the same reason.

Adding .lower() only widens acceptance to True and TRUE, so no value that works today changes meaning.

Fixes #4240

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline, Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link to it if that's the case: ACCELERATE_BYPASS_DEVICE_MAP is the only boolean env flag read case-sensitively, so =True is silently ignored #4240
• Did you make sure to update the documentation with your changes? The variable is undocumented today; I have not added docs for it here, but I am glad to if you would like it listed with the other environment variables.
• Did you write any new necessary tests?

On that last box, deliberately unticked: the three call sites need a distributed Accelerator and a model loaded with device_map='auto', so a faithful test is heavier than the change. Happy to add one if you would rather have it, and equally happy to route these flags through str_to_bool instead, which would also make them accept 1, yes and on like the library's own documented parser does. I kept that out of here because it is a wider blast radius than a bug fix.

Who can review?

Anyone in the community is free to review the PR once the tests have passed.