cargo publish should warn if it fails to detect a README

[cbiffle]

Problem

I often cargo publish an initial version of a crate without a README... by accident. This is because my naming convention for README files doesn't match what Cargo expects. (I'll open a second bug to ask about whether my convention is common enough to support, but, that's a different matter.)

I always run cargo publish -n first and check for warnings. Cargo is very consistent about warning about missing repository, documentation, license, and other fields, but does not warn if it can't find a README file and none is specified in the Cargo.toml.

Proposed Solution

I'd like to include the readme in the list of fields that cargo publish warns about. (This list appears to be specific to publish, from the source code.) It would be important for this check to be performed after readme autodetection, so that users who have a README named README.md don't get hassled --- only those of us with README.somethingelse.

If there's a real use case for publishing (not packaging!) crates without a README, and people want to do that without a warning, then this might not make sense without a way to suppress the warning.

Notes

In case you're curious, my convention is README.mkdn. But that's not really material for this report.

[weihanglo]

According to our current doc, readme is encouraged to be filled in. Cargo would also auto-detect from some paths when missing.

Based on these facts, if we are not going to change the recommendation, I feel like it is reasonable to add README to metadata check.

[epage]

The initial warning was for description and license (#902, #903). We later extended this to repository, homepage, and documentation (#998, #1053). In all of those discussion, I never saw any talk of README, for or against. Nor am I finding any other issues about this. I'm a bit surprised at that.

Generally, we take the stance that we shouldn't emit warnings without a reasonable way to turn them off until we have #12235. We didn't take that approach with those prior warnings, so maybe thats fine here. One way to soften this is to only warn if the field is None so people can set package.readme = false to disable it.

[cbiffle]

Ooooh, I like package.readme = false as a mechanism for explicitly stating that no, thank you, I really don't want to have a readme.

[weihanglo]

We didn't take that approach with those prior warnings, so maybe thats fine here

We have --no-metdata on cargo package but not on cargo publish. Wonder if we should also replicate it to cargo publish.

[epage]

I forget that --no-metadata exists. I feel like #12235 is close enough that we should focus on that.