Unbox writer (3-5% faster)

[alamb]

• Part of Improve single threaded performance #6

Rationale

Using a Box<dyn Write> when there is only one actual writer implementation prevents certain inlining from occurring.

I noticed this while looking at some performance traces

Changes

Use the actual types for the writer rather than Box<dyn Write>

I measured a small but measurable 3-5% improvement using time target/release/tpchgen-cli -s 1 --output-dir=/tmp/tpchdbgen-rs

branch	time
main	0m8.573s
this branch	0m8.297s

[clflushopt]

LGTM fyi I did experiment with different buffer capacities for the BufWriter I noticed there are speed ups when I increased the buffer to 16 or 32 MBs on my Linux machine but I am not confident it makes sense since it would depend on page sizes and OS buffer caches because the improvements were not noticeable on my Macbook so i didn't add those.