fix frame_clone_and_push_hdr: guard stack overflow at the push site

[jeso-mchp]

Revert the argc > FRAME_STACK_MAX check added in db43211 ("Frame stack size check added") and replace it with a proper guard in frame_clone_and_push_hdr().

Problem

The check in argc_frame() compares argc (total remaining argv) against FRAME_STACK_MAX (the per-frame header stack depth). These are fundamentally different quantities, argc includes arguments for ALL subsequent tx/rx clauses, not just the current frame being parsed.

A single tx clause like tx eth0 eth smac ::1 dmac ::2 consumes 7 argv entries but only pushes 1 header onto the frame stack. With FRAME_STACK_MAX=200, just 29 tx clauses (29 × 7 = 203 args) trip the check even though each individual frame only uses a few stack entries.

This breaks real-world usage. For example, the switchdev test framework's FDB stress test batches 100 tx clauses into a single ef invocation:

smacs.each_slice(100).each do |chunk|
cmd = 'ef'
chunk.each do |smac|
        cmd += " tx #{$pp[1]} eth smac #{smac} dmac #{dmac}"
end
$ts.pc.run cmd
end

This fails due to this argc check bug.

Fix

Remove the argc check from argc_frame() and guard the actual array push in frame_clone_and_push_hdr() where f->stack[] is written.

frame_clone_and_push_hdr() is the only place that writes to f->stack[], so this single guard is sufficient.