My criterion for delayed imports is that they're only worth it if the
majority of users of the module would benefit from it, otherwise you're
just moving latency around unpredictably.

mktime_tz is not used anywhere in the standard library and grep.app
indicates it's not got much use in the ecosystem either.

Distribution.files is not nearly as widely used as other
importlib.metadata APIs, so we defer the csv import.

Before:
```
λ hyperfine -w 8 './python -c "import importlib.metadata"'
Benchmark 1: ./python -c "import importlib.metadata"
  Time (mean ± σ):      65.1 ms ±   0.5 ms    [User: 55.3 ms, System: 9.8 ms]
  Range (min … max):    64.4 ms …  66.4 ms    44 runs
```

After:
```
λ hyperfine -w 8 './python -c "import importlib.metadata"'
Benchmark 1: ./python -c "import importlib.metadata"
  Time (mean ± σ):      62.0 ms ±   0.3 ms    [User: 52.5 ms, System: 9.6 ms]
  Range (min … max):    61.3 ms …  62.8 ms    46 runs
```

for about a 3ms saving with warm disk cache, maybe 7-11ms with cold disk
cache.