Improve scheduler E2E test code

[rockbmb]

In #446 , there were spurious failures with the scheduler E2E test suite because of unrobust code patterns in tests.

Example job: https://github.com/open-web3-stack/polkadot-ecosystem-tests/actions/runs/19148571343/job/54732595597

Example of offending code:

polkadot-ecosystem-tests/packages/shared/src/scheduler.ts

Lines 289 to 311 in 9362360

	const scheduleTx = client.api.tx.scheduler.schedule(targetBlockNumber!, null, 0, adjustIssuanceTx)
	await scheduleInlineCallWithOrigin(client, scheduleTx.method.toHex(), { system: 'Root' }, testConfig.blockProvider)
	const oldTotalIssuance = await client.api.query.balances.totalIssuance()
	await client.dev.newBlock()
	let scheduled = await client.api.query.scheduler.agenda(targetBlockNumber!)
	expect(scheduled.length).toBe(1)
	expect(scheduled[0].isSome).toBeTruthy()
	await check(scheduled[0].unwrap()).toMatchObject({
	maybeId: null,
	priority: 0,
	call: { inline: adjustIssuanceTx.method.toHex() },
	maybePeriodic: null,
	origin: {
	system: {
	root: null,
	},
	},
	})

Using client.dev.setStorage, the scheduler's agenda for a given block is manually modified to execute a certain task with a desired origin.
This block's prior agenda, from the live network, is erased, so test code executed normally.

The issue occurs when the executed call is itself a scheduling call, and the block in which it is scheduled already contains tasks, which schedule.schedule(Named) will not, as expected, delete.

This makes the above expects to fail, and the test to fail, incorrectly.

This issue is about correcting these unsafe accesses to the scheduler's agenda in tests, replacing them with a proper search.