feat(panel): tell every user what the in-memory cache costs, at the top of Session Settings

The previous version only spoke up on machines with 36 GB or less, and it
spoke in the vocabulary of someone who already knows what a paged KV cache
is. Both were wrong. Anyone deciding how to spend their memory deserves the
trade-off, and the people who most need it are the ones who would not
recognise the term.

It now appears unconditionally at the top of Session Settings, above Server
Settings, in plain language: the cache keeps recent prompts in RAM to answer
follow-ups faster, it uses extra memory to do that, and switching it off is
only about 2 percent slower while giving that memory back to the model. It
then names the exact control to change and tells them to leave the SSD cache
on so prompts are still remembered.

Shipped in all five locales rather than English-only. Tests assert every
locale carries it, names the setting, quotes the 2 percent rather than
hand-waving, and never reads as though the app changed anything on the
user's behalf.

The RAM detection added for the gated version is removed along with it; no
threshold decides who sees this any more.