Provide Amount & co in no-alloc

[Kixunil]

Using the crate without allocation was previously disabled making the
crate empty without the feature. This chage makes it more fine-grained:
it only disables string and float conversions which use allocator. We
could later provide float conversions by using a sufficiently-long
ArrayString.

Note that this is API-breaking because we disallow calling the methods of the sealed SerdeAmount trait. However I think it should've been obvious that the thing is internal and calling them is not a great idea. (BTW I only learned this trick recently).

Closes #2389

[coveralls]

Pull Request Test Coverage Report for Build 7678592996

• -8 of 45 (82.22%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.001%) to 84.308%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
units/src/amount.rs	37	45	82.22%

Totals
Change from base Build 7668596887:	0.001%
Covered Lines:	19315
Relevant Lines:	22910

---

💛 - Coveralls

[apoelstra]

Fun! I also learned this trick recently (maybe from a cargo-semver-checks blog post).

The use of #[no_std] is new to me ... a long time ago I knew about this attribute but I don't know what it does.

[apoelstra]

ACK ac26171 maaybe private should be renamed to internal_api or something

[Kixunil]

maybe from a cargo-semver-checks blog post

That's definitely where I found it.

The use of #[no_std] is new to me ... a long time ago I knew about this attribute but I don't know what it does.

Without it each crate has an implicit extern crate std injected with the std prelude etc. The attribute turns it off, so then you need to manually import extern crate std if you want and import prelude items. One approach is to do exactly that in no_std libraries with std enabled, the other is to use not(std) on the attribute and a bunch of other places.

Personally I find it easier to reason about additive features. When something requires feature x just slap feature = "x" there and it's done. Also seeing #[no_std] immediately, not somewhere between other attributes is nice for discoverability.

[Kixunil]

maaybe private should be renamed to internal_api or something

Usually it's called sealed but I didn't think it's important to change the name.

[apoelstra]

Usually it's called sealed but I didn't think it's important to change the name.

Well, by definition we can rename it without breaking the API so I'm not gonna hold up this PR over it :). But if in the future somebody suggests a clearly-better or well-established name we should move to it.

[tcharding]

ACK ac26171