Apply obvious fixes for issues identified during initial sweep run

[SeanNijjar]

Ticket

Link to Github Issue

Problem description

During an initial run through of the n300 allgather sweep tests, a handful of "obvious" errors were identified. This PR resolves those obvious issues

What's changed

A few changes:

• Relaxed test case invalidation at pytest level that would be overly conservative and skip some cases because it thought they couldn't fit in L1/DRAM (even though they could)
• Updated timeout for sweeps as it was too long (after reset, device initialization seemed to take a while, often time leading to another timeout before the main test body was even enterred
• Fixed all-gather edm buffer sizing to support (much) larger pages.

Checklist

• Post commit CI passes: https://github.com/tenstorrent/tt-metal/actions/runs/11170239688
• t3k pipelines: frequent, nightly, model-perf: https://github.com/tenstorrent/tt-metal/actions/runs/11170251637
• tg pipelines: https://github.com/tenstorrent/tt-metal/actions/runs/11170262941
• N/A Blackhole Post commit (if applicable)
• (above) Model regression CI testing passes (if applicable)
• N/A Device performance regression CI testing passes (if applicable)
• N/A New/Existing tests provide coverage for changes