Use musl implementation of strftime rather than custom JS code. NFC

[sbc100]

This leads to some code size savings in several tests, reduces the differences between standalone mode and non-standalone mode and reduces the amount of custom JS code we need to maitain.

In simple microbenchmark that does nothing but export strftime we do code size win at -O0 but a regression in codesize at -O2:

$ wc -c old_O0.*
74026 old.js
12124 old.wasm
86150 total
$ wc -c new_O0.*
62120 new.js
22852 new.wasm
84972 total

$ wc -c old_O2.*
8492 old.js
 194 old.wasm
8686 total
$ wc -c new_O2.*
 5040 new.js
14291 new.wasm
19331 total

This regression comes from the fact that snprintf is used in the implementation of strftime and its hard to optimize away the cost of that function.

In practice I suspect that any application large enough to call strftime is likely already directly or indirectly pulling in snprintf so I would hope that this will be codesize win in practice since it removes about 3k of JS.

[kripken]

Agree that likely in real-world cases this shrinks code size, and is a very nice simplification.

[dschuff]

Agreed, I think this is worth it.

[sbc100]

I had some second thoughts here because I discovered that C++ (specifically C++ i/o streams) end up depending on strftime. So this could effect more programs that I initially expected.