Remove redundant reserve call

[braddunbar]

Note: Miri is currently failing in CI, but is fixed by #673

We're calling reserve twice because put_u8 already calls it.

1. put_u8 calls put_slice
2. BytesMut::put_slice calls BytesMut::extend_from_slice
3. BytesMut::extend_from_slice immediately calls reserve

[taiki-e]

I'm not sure what this TODO comment originally meant, but if it means something like "reserve (in here or put_u8) should be able to omit if the compiler knows the actual length of iter" it may not have been resolved yet.

(In my experience, the compiler optimizations around the reserve have not worked that well.)

EDIT: but seanmonstar who wrote this TODO comment approved this PR, so this is probably not the meaning I have imagined here.

[braddunbar]

Thanks @taiki-e! I think it refers to #313, which was just after the introduction of the comment in #305. I should've mentioned that in my commit though. Thanks for taking a look!

[braddunbar]

Updated with #673 to fix miri tests!

[Darksonn]

Before accepting this PR, I would like to see a test that calls extend with an iterator containing more items than the lower limit of the size hint. I'm worried about the panic conditions of put_u8.

Edit: I guess the behavior is probably okay due to how BytesMut pretends to have an infinite capacity? Still, the test would be good to add to guard against future changes.

[braddunbar]

@Darksonn Sure thing! Added in 76dae7b. I think the following test technically covers it via Extend<&'a u8>, but I like being explicit.