Some slightly performance & style updates

[yhx-12243]

• In the function fn as_bytes in impl CompactString, there is an redundant slicing (with a self.len() call) which is unnecessary after the update c95edd5.
• This crate has documented that for 64-bit architectures the capacity will be always inlined. So the logic of is_heap can be replaced with directly returning false to reduce some code generation.

[yhx-12243]

It seems that the test needs to be changed, because it's probably that the compiler optimized out the whole test and did not allocate memory at all.

See https://rust.godbolt.org/z/qc349PrGv for a shorter analogue.

[yhx-12243]

May be we should wrap this test in a something like core::hint::black_box, or maybe do other things to make sure the compiler will not optimize out the "unused" allocation of long string.

[yhx-12243]

Hello @ParkMyCar , do you have time to review this PR if you're spared?

[ParkMyCar]

Geez @yhx-12243 this sat for so long, I'm sorry about that! Thanks for the PR and the performance improvements, they look good to me!