Formatting for float and integer columns via Intl.NumberFormat

[texodus]

Adds formatting options for integer and float columns. From #2539 which this PR supersedes:

This PR aims to create a universal number string formatter using the Intl.NumberFormat() constructor as the base datatype. This PR creates a column settings control on the viewer and integrates it into the Datagrid. Further work will integrate this control into the D3FC plugins and the OpenLayers plugin.

Things I changed from the predecessor PRs:

• Redesign of the enable/disable paradigm for settings, switch to default/non-default.
• Graphical redesign.
• Converted radio lists to <select>
• Fixed examples which still use columns.
• Renamed column_config to columns_config for consistency (this is the corollary for the columns property and plugin/plugin_config).
• Removed update_and_render(). This method is confusingly named - it does not call update(), it updates the VIew config (which is very expensive) and re-renders. The View config diffs and drops spurious updates, but it is still not behavior we should rely on, plus there is a better option in Renderer::update().
• Refactoring/bugfixes/etc.

There are a bunch of issues remaining with this PR that I feel need to be resolved:

• columns_config (was column_config from columns) is still a weird term. The suffix config is autological and fails to describe what distinguishes this field from its sibling columns.
• Moving this property from plugin_config is not obviously better to my eye either. At least as a member of plugin_config, this kept the save()/restore() API for plugins simple. Additionally, this has been moved to share with e.g. group_by, which is another config depth we should conflate less (as some of these properties to to the View() constructor, but some must be removed).
• Number formatting options still have some unexpected behavior. I like the pattern of having the UI update to resolve conflicts, but there are a lot of cases to consider.

IMO we should be unwinding this ambiguity completely, not shuffling it around. This:

{
    version: "2.7.1",
    plugin: "Datagrid",
    plugin_config: {
        columns: {
            datetime: {
                format: "custom",
                timeZone: "America/Curacao",
                datetime_color_mode: "foreground",
            },
            "Order ID": {
                format: "link",
                string_color_mode: "foreground",
                color: "#ff0000",
            },
        },
        editable: true,
        scroll_lock: false,
    },
    columns: ["Order ID", "datetime"],
    expressions: { datetime: "datetime(100)" },
}

... is now this ...:

{
    version: "2.7.1",
    plugin: "Datagrid",
    columns_config: {
        datetime: {
            format: "custom",
            timeZone: "America/Curacao",
            datetime_color_mode: "foreground",
        },
        "Order ID": {
            format: "link",
            string_color_mode: "foreground",
            color: "#ff0000",
        },
    },
    plugin_config: {
        editable: true,
        scroll_lock: false,
    },
    columns: ["Order ID", "datetime"],
    expressions: { datetime: "datetime(100)" },
}

... but IMO this makes more sense, and mirrors how this data structure is ultimately dissected during restore():

{
    version: "2.7.1",
    plugin: "Datagrid",
    style: {
        editable: true,
        scroll_lock: false,
        columns: {
            datetime: {
                format: "custom",
                timeZone: "America/Curacao",
                datetime_color_mode: "foreground",
            },
            "Order ID": {
                format: "link",
                string_color_mode: "foreground",
                color: "#ff0000",
            },
        },
    },
    view: {
        columns: ["Order ID", "datetime"],
        expressions: { datetime: "datetime(100)" },
    }
}

Refactorings like this are costly on the Perspective community (and me), but this would be complex to do today as plugin_config is untyped and the state is managed externally, so I can accept this conceit to the structure of the code for now. It is still a fairly significant breaking change to the persistence API.