fix(openai): correct rendered-context framing for annotation-only tool imaging

[teamchong]

Salvages the still-valid part of #19 (by @rldyourmnd) and rebases it onto current main.

Why

main already fixed the double-billing that #19 originally targeted: commit 4c78363 reworked the GPT path to strip parameter annotations from the native tools[] and image only those annotations (schemaAnnotationLines), keeping each tool's name and top-level description native. That superseded #19's functional change.

What it left behind is stale framing text: CHAT_HEADER / RESPONSES_HEADER and the two pointers still tell the model the images hold "full tool/schema documentation" and that the native JSON is merely supplemental — no longer true now that only parameter annotations are imaged.

What

• Reword the four framing constants so they match what is actually imaged: the native JSON tool definitions carry each tool's name and description and stay authoritative; the imaged parameter annotations are supplemental.
• Export the constants and add a regression test (rendered-context wording) asserting the framing never claims "full tool" docs and always points at the native definitions.
• No behavior change to token accounting or imaging — framing text only.

Credit

The rendered-context wording work is @rldyourmnd's from #19; the commit here is authored under their name. This PR only rebases it onto main's annotation-only design and updates the wording to describe that design accurately.

Supersedes #19 (whose functional change is already in main via 4c78363) — #19 can be closed.

Full test suite green (783 tests) + typecheck clean.