Trial implementation of fix for FLUID-4681

[Antranig Basman]

A user report in #fluid-work today revealed a problem with our rich text integration with more recent 
versions of TinyMCE than the one in our framework demo and implementation. The current version is 3.4.8 
(actually now 3.4.9), as opposed to the 3.2.3.1 which we last tested with. Since then, a serious problem has 
arisen which makes the integration unusable on the current FireFox (12) - the first time editing is 
triggered by any means, the component will immediately finish the edit operation and fold up. The issue 
doesn't seem to appear on other browsers.

Detailed description at http://issues.fluidproject.org/browse/FLUID-4681

I've issued an interim pull request which shows the issue resolved but it is not yet a serious request for a 
pull - see https://github.com/fluid-project/infusion/pull/215

In particular, before this can be pulled back in, we need to
i) Strip back out the version of TinyMCE I have committed for testing, and put it up into our CDN with the 
others,
ii) Get design input on whether the inconsistency of the behaviour of this component with the simple 
InlineEdit (that it cancels an edit on blur, rather than committing it) is really correct, and also what the 
correct behaviour should be on receiving a focus outside the component other than on click
iii) Determine whether the new "facade API" for deadMansBlur is really reasonable given its "stateful model" 
for its lifetime using the "reArm", "proceed" etc. methods which were not used by the previous stateless 
system which did not defer to globalDismissal.

In the meantime, Nick should be able to make a build from it and use it for interim work. Testing showed 
that the issue is purely caused by the difference in TinyMCE version, the version of jQuery is unimportant.

Cheers,
A

NB - since our last testing, there has been a further regression in the component in that our "HTML 
canonicaliser" no longer functions correctly for the current rich text components and browser combinations. 
Both the integrations fail to correctly round-trip markup, which means that on triggering an edit and making 
no changes, the "Undo" state incorrectly becomes set.