Use "shorter representation" codepath for 64-bit integers

[hikari-no-yume]

Fixes a lot of cases of (uint64_t)UINT64_C(0) that should just be (uint64_t)0.

Some examples from a small program of mine:

<   _3 = llvm_lshr_u64(_1, UINT64_C(1));
---
>   _3 = llvm_lshr_u64(_1, 1);
90c90
<     _4__PHI_TEMPORARY = UINT64_C(0);   /* for PHI node */
---
>     _4__PHI_TEMPORARY = 0;   /* for PHI node */
99c99
<   _5 = llvm_add_u64(_4, UINT64_C(1));
---
>   _5 = llvm_add_u64(_4, 1);
157c157
<   uint32_t _23 = puts(((&str.array[((int64_t)UINT64_C(0))])));
---
>   uint32_t _23 = puts(((&str.array[((int64_t)0)])));
162c162
<   uint8_t* _25 = *((&_19[((int64_t)UINT64_C(1))]));
---
>   uint8_t* _25 = *((&_19[((int64_t)1)]));
165,166c165,166
<   uint32_t _28 = printf(((&_OC_str_OC_1.array[((int64_t)UINT64_C(0))])), _25, (((uint32_t)(bool)_27)));
<   uint8_t* _29 = *((&_19[((int64_t)UINT64_C(2))]));
---
>   uint32_t _28 = printf(((&_OC_str_OC_1.array[((int64_t)0)])), _25, (((uint32_t)(bool)_27)));
>   uint8_t* _29 = *((&_19[((int64_t)2)]));
169c169
<   uint32_t _32 = printf(((&_OC_str_OC_2.array[((int64_t)UINT64_C(0))])), _30, _31);
---
>   uint32_t _32 = printf(((&_OC_str_OC_2.array[((int64_t)0)])), _30, _31);

The condition I am changing was introduced by https://github.com/JuliaComputingOSS/llvm-cbe/commit/7da56e67e8505116aed007a45f467d84c96127e2.