testscript: add build:tag condition

[twpayne]

As discussed with @mvdan in #149.

[twpayne]

The build: prefix is needed to distinguish between set and not-set tags. If the syntax is only tag (e.g. [go1.19]) then it is not possible to distinguish between a not-set tag and an unknown condition.

[rogpeppe]

Sorry for the slow review. This LGTM modulo a couple of minor suggestions, and I think it supercedes #149, right? Thanks for proposing it!

[twpayne]

Thanks for the review and zero stress about the review time - you're giving the gift of open source so there are no expectations or obligations.

I've applied your suggested changes and rebased on the latest master, so this should be good to merge.

[twpayne]

Anything else needed here?

[mvdan]

Apologies for my delay here :) Two thoughts:

1. I still think that we should just move this part of gotooltest into the main testscript package, so that conditions like [go1.17] or [gc] are supported by all test scripts. Like you say, this can be useful for any program written in Go, even if it doesn't interact with the Go toolchain and doesn't need gotooltest. Supporting both [go1.17] and [build:go1.17] seems unnecessary.

2. Similar to the above, I don't really agree with adding [build:buildtag] support when we in fact don't support arbitrary build tags. At best, it's unnecessary, and we should stick with [go1.17] as per the above. But I'm also worried about misleading users into thinking that e.g. [build:purego] will work when it won't.

[twpayne]

#154 implements @mvdan's suggestions, superseding this PR.

[twpayne]

Closing in favor of #154.