[css-tables] correct content measurement based on test findings

[gregwhitworth]

While trying to verify the content measurement we found that the spec text is not correct and needs to be re-written for colgroup widths.

• the outer min-content width of the column group, if any
• the outer min-content width of the column, if any
• the outer min-content width of each cell in the column whose column-span is 1 (or just the one in the first row in fixed mode) or 0 if there is none

For example in this test based on the above spec text both columns should result in 51px and this is not what any browser does.

<table cellspacing="0" cellpadding="0">
    <colgroup style="width: 51px;">
        <col style="width: 20px;"></col>
        <col style="width: auto;"></col>
    </colgroup>
    <tr>
        <td style="width: 10px;"></td>
        <td id="four" style="width: 20px;"></td>
    </tr>
</table>

[gregwhitworth]

This is related to issue #1022 but it's more broken than we thought :)

[gregwhitworth]

@FremyCompany While I think this should be an obvious resolution since no browser behaves this way, this is a technical edit and we should probably ensure we don't need to resolve on it. @atanassov @astearns What is the process in the case that we make the spec conform to reality of all UAs?