output: memoize calculate_timekey to reduce redundant calculations (#5288)

**Which issue(s) this PR fixes**: 
Fixes #

**What this PR does / why we need it**: 
`calculate_timekey` is called for every emitted record.
Since logs usually arrive chronologically, consecutive records almost
always fall into the same timekey.
However, the current implementation performs `%` and `to_i` for every
single call.

When running `rake benchmark:run:in_tail` with 10 GB of data, this
memoization improved the overall throughput:

| Benchmark | before | after |
|---|---|---|
| `rake benchmark:run:in_tail` | 43.170137 | 41.584905 |

**Docs Changes**:
N/A

**Release Note**:
* output: Improve performance by memoizing calculate_timekey

Signed-off-by: Shizuo Fujita <fujita@clear-code.com>