fix(rune): address conductor migration nitpicks

[11suixing11]

Summary

• validate the rune channel before calling the base plugin constructor
• make rune initialisation idempotent so primitive rune exports are not duplicated
• load the host tab by the shared Rune tab name instead of the first tab entry
• share rune deserialisation through the bundle protocol and cover serialize/deserialize round trips

Tests

• git diff --check
• corepack yarn workspace @sourceacademy/bundle-rune test
• corepack yarn workspace @sourceacademy/bundle-rune tsc
• corepack yarn workspace @sourceacademy/bundle-rune lint
• corepack yarn workspace @sourceacademy/bundle-rune compile
• corepack yarn workspace @sourceacademy/tab-Rune tsc
• corepack yarn workspace @sourceacademy/tab-Rune lint
• corepack yarn workspace @sourceacademy/bundle-rune build
• corepack yarn workspace @sourceacademy/tab-Rune build

Not run locally

• corepack yarn workspace @sourceacademy/tab-Rune test: Playwright 1.59.1 rejects Chromium installation on ubuntu26.04-x64 in this environment

Addresses #827

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[AaravMalani]

Are there instances where an error is encountered without this change?

[11suixing11]

No observed user-facing runtime failure from this constructor ordering specifically. This one is a defensive/parity fix from the linked issue: with the old order, super(conduit, [runeChannel], evaluator) receives [undefined] before Rune gets a chance to throw its own "Rune channel is required" error. Moving the guard before super() keeps the invalid channel out of BaseModulePlugin entirely.

[AaravMalani]

Same here, I'd love to hear the motivation behind this PR! Do you have a code sample out of curiosity?

[AaravMalani]

Ah, I just noticed the attached issue, let me get home and review this

[11suixing11]

Quick concrete repro for the initialise() part: create a RuneModulePlugin, call await plugin.initialise(), then call it again. Before this change, the second call re-runs super.initialise() and re-pushes the exported methods/primitive runes, so plugin.exports.map(each => each.symbol) grows duplicates. The added initialise only exports primitive runes once test pins that behavior. The tab lookup part is similarly low-risk cleanup: createRunePlugin(["Other Tab", RUNE_TAB_NAME]) should load RUNE_TAB_NAME, not whatever happens to be at tabs[0].