Cleanup client IDs from the registered checkbox map when the components are removed from DOM.

Tested with ICEfaces 4 trunk r49407, Tomcat 8, IE 11, FF 47 Chrome 54
Issues found:
1.) ace:radioButton apps that are not using a dataTable or repeat structure are not functioning properly. mutuallyExclusive = true but is not respected.
AN example of this can be found in the showcase ace:radioButton > Overview and Custom Style apps.
To reproduce:
Open either of the radioButton showcase demos and select the unselected radioButton. Both buttons stay selected. After they are both selected you can select one to force it to deselect (not proper behaviour). After that they become mutually exclusive.

2.) ace:radioButtons no longer retain their itemValue once selected. They return false instead. This can be seen in the showcase ace:radioButtons > Overview.
When first loading the demo radioButton One is selected, the value shown is "The selected radios are: One". After selecting any other radioButton the value shown is "The selected radios are: false". This issue is seen whether or not the radioButtons are inside a dataTable or repeat structure.

3.) Showcase ace:checkboxButtons > all demos fail. No buttons can be selected. There are console errors when loading the demo
Chrome: coalesced.js.jsf?ln=ice.core&dgst=kbiglm:14029 Uncaught TypeError: Cannot read property 'init' of undefined(…)
FF:
TypeError: a.dataset is undefined

1) Fixed ice.ace.registerLazyComponent to return the widget reference when just created.

2) Modified CheckboxButtonsRenderer to replicate what scripts are rendered by CheckboxButtonRenderer after the recent changes. Same was done for RadioButtonsRenderer. I didn't realize how little code is reussed by the buttons renderers.

3) Changed RadioButtonsRenderer to avoid using the inherited RadioButtonRenderer.getConvertedValue method since the value is always cast to a boolean. This issue was actually a regression introduced by ICE-11153 refactoring.

Here are the optimisation results. The profiling was run during the switching from 10 to 1000 rows in showcase > buttonGroup > Group In Column demo.

Before optimization the measured interval (using visual clues) was 8 seconds. The profiling shows that our JS scripts took about 5 seconds.

After optimization the measured interval (using visual cues) was 2 seconds. The profiling show that only 890 milliseconds was used by JS, 360 milliseconds was the actual time used by our code.

1) I don't see any difference in behaviour from EE 4.1.0 GA release. One still needs to click on "Submit" button (after "Clear" button is clicked) to actually clear the outputText that renders the selected item. This is needed because the "Clear" button does not issue a submit, it just clears the radio buttons locally.

2) This is again a regression caused by ICE-11153 commits. The fix restores RadioButtonsRenderer.renderResetSettings method to properly invoke the ice.ace.radionbutton.clear function for multiple selection (instead of relying on the inherited method meant for RadioButton).