Only get draw_target-width when we actually draw

[jaheba]

When calling BarState::draw, there is a call to self.draw_target.width().

However, that operation can be expensive and the width is not needed, if there is no draw-target.

Thus, this PR delays the calling .width() if there is not target to draw to.

For comparison, I've used this minibenchmark:

    let bar = ProgressBar::new_spinner();

    for n in 0..1_000_000 {
        bar.set_message(format!("{}", n));
    }
    bar.finish();

Using the latest release it runs in about 400ms, whilst the fixed version uses about 100ms.

The issue surfaced with uv, when resolving many packages: astral-sh/uv#10384

[jaheba]

@djc Tests pass locally now.

[jaheba]

Actually, this is not working, since self.draw_target.drawable(...) will set the time of latest update, and then second call to .drawable can return None.

[jaheba]

Should (hopefully) work now.

The contrast in speed is actually bigger when using hyperfine --show-output. The fixed version takes ~120ms, while the current version takes ~700ms.

[jaheba]

@djc Let me know if you need any else for this PR.

[djc]

LGTM modulo some small nits.

Would be nice if you can squash your commits.

Thanks for these performance wins. Do you have more in the pipeline? Would you like a release once these are merged?

[djc]

Nit: please avoid unrelated cosmetic changes.

[djc]

Nit: please use Self instead of Drawable here.

[jaheba]

Thanks for the review!

I've addressed your comments.

I don't have any more changes (besides #684) planned. I wanted to look into some basic benchmarking to see if there are any other possible performance improvements.