API: Differentiate bad TXT update error.

[cpu]

Previous to this commit, if the update message had a valid subdomain but
an invalid TXT value the error returned was for a bad subdomain. This can
confuse developers who were POSTing junk TXT records to test acme-dns 😊
Only TXT values that are exactly 43 chars are considered valid:

acme-dns/validation.go

Lines 37 to 40 in 830cceb

	if utf8.RuneCountInString(s) == 43 && utf8.RuneCountInString(sn) == 43 {
	// 43 chars is the current LE auth key size, but not limited / defined by ACME
	return true
	}

This commit adjusts the webUpdatePost error handling such that
!validSubdomain(input) and !validTXT(input) give distinct errors.

The !validSubdomain case should never happen in webUpdatePost
because auth.go's Auth function already vets the post data
subdomain but I retained the error handling code just in case.

Unit tests for an update with an invalid subdomain and an update with an
invalid TXT are included.

I did my best to match the repo style for the code/tests but I admit
to hacking this together quickly without having read most of the codebase
yet! Please critique liberally.

[coveralls]

Coverage decreased (-0.5%) to 91.946% when pulling 92f8cc2 on cpu:cpu-differentiate-errors into 978ac5d on joohoi:master.

[joohoi]

Thanks for the PR, this certainly adds clarity to the source of error for people poking around and evaluating acme-dns for use.

I did my best to match the repo style for the code/tests but I admit
to hacking this together quickly without having read most of the codebase
yet! Please critique liberally.

It's all good!

[cpu]

Thanks @joohoi, happy to help!