fix(app): put the transmit pop under the receptions it sat over

[khagele]

Closes #602

With sound on, auto-discover was the loudest thing in the mix. A sweep of N targets is N + 2 pops every ten seconds, each at gain 0.16 with a 160 ms tail through the whole reverb. A direct network dit at -110 dBm peaks at 0.078, so the cue for "a frame went out" was 2.1 times over the cue for "something was heard".

In this PR

TX_POP in sound.js holds the levels Kasper chose by ear on 30 September, in a lab page running the real engine with the pop adjustable:

	was	is
peak gain	0.16	0.05
tail	160 ms	70 ms
share through the master and its reverb send	100%	15%, the rest dry at the same level

• The dry share goes through a second lowpass into the same output gain, so it gets the master's rounding without its reverb send.
• cuePeak is the level cue() plays a direct dit at, pulled out so the pop can be held against it.
• Every transmission still sounds: two pops for the Discover broadcast, one per trace-ping. Sounding only the Discover pops (direction 2) and a silent auto-discover (direction 3) were both in the lab and not chosen.
• The rule of App: Discover FAB pulses only for the broadcast, not for trace-ping transmissions #254 is untouched: the cue follows the frame.
• docs/2026-07-16-sound-modes.md carries the new levels as an amendment, and there is a changelog entry.

Acceptance

A direct network dit at -110 dBm (0.078) is 1.6 times over the pop (0.05). Measured in the lab at the master, behind the 0.9 output gain: pop 0.042 to 0.045, network dits 0.064 to 0.079, trace replies 0.143 to 0.155 (taken with the reverb share at 0; the dry level is the same at 15%).

Verification

• soundengine.test.js, 3 new tests: every family's dit at -110 dBm peaks over the pop; cue() plays a dit at cuePeak; a pop reaches the reverb at TX_POP.reverb of a dit's share and the output at the same dry level.
• Each was run with its change reverted: 5 mutations, 5 red.
• app: 1626 tests green, eslint clean, build clean. web: 762 green (the changelog copies).
• The levels were heard in the lab page, not in the app build on a phone.

🤖 Generated with Claude Code