feat: iris_execute use persistant session

[cwennerh]

It would be very useful if running commands via iris_execute could exist in a persistent session such that you could create some object, set properties and run a method with it. Then, after viewing the output, the object would still exist, so claude could update that same object and continue running methods with it.

[isc-tdyar]

This is on our radar. The main constraint is that iris_execute runs each call in a fresh
objectgenerator context — there's no process continuity between calls.

The most practical path forward is a global-backed context store keyed on the MCP session:
%Persistent objects get saved and their IDs stored in a session global between calls, then
reloaded at the start of the next call. This covers the main use case (building/modifying
persistent domain objects across turns).

Pure in-memory state (non-%Persistent objects, local variables) can't survive between
calls by nature — that's an IRIS constraint we can't work around without a permanently
live iris session process per connection, which brings its own complexity (cleanup on
disconnect, Windows compatibility, multiplexing).

If the persistent-object pattern would meet your needs, we can spec it out. What's your
primary use case — domain object manipulation, or something more procedural?