docs: guide readers between JLedGroup and JLedRefGroup, drop deferred-feature promise

[jandelgado]

Documentation only, no code changes.

What

1. New "When to use what?" section in the group chapter, so readers can pick a group type without reading both subsections first. It leads with JLedGroup as the default and frames JLedRefGroup around the decision that drives the choice: whether you already keep your LEDs as named variables.

In that case JLedRefGroup is substantially more memory efficient, because a JLedGroup stores its own copy of every LED, so a named LED ends up stored twice. Measured on atmega328p at -Os, six named JLed objects plus the group:

6 named JLed	flash	RAM
JLedGroup	3096	358
JLedRefGroup	2888	198

Per element that is 54 bytes (24 for your named object + 30 for the JLedAny copy) against 28 (24 + a 4-byte JLedRef).

The section states this as a conditional benefit rather than a blanket one, because it does not hold when the LEDs live only inside the group and you never touch them individually. There the two are close to a wash, and the sign even flips with LED type, since sizeof(JLedRef) + sizeof(JLedHD) (32) exceeds sizeof(JLedAny) (30):

6 LEDs, group-only	JLedGroup RAM	JLedRefGroup RAM
JLed	214	198
JLedHD	214	222

2. Removed two forward-looking promises about per-element events:

• "Per-element events (reacting to individual LEDs within a group) are not yet supported."
• "Full-depth, per-leaf visibility is planned via a future per-element callback."

That feature is deferred with no timeline, so the README should not advertise it.

3. Documented what works today in its place. To react to a single LED inside a group, keep it as a named variable and watch its own state across the Update() call:

bool was_running = led1.IsRunning();
group.Update();
if (was_running && !led1.IsRunning()) {
    Serial.println("led1 finished");
}

Verified against the host test mocks: this fires exactly once, on the tick led1 completes, while other group members keep running.

Notes

• Table of contents updated; all internal anchors resolve.
• One pre-existing broken link, [Percentage](#percentage) at line 644, is left alone to keep this PR focused. There is no matching heading in the file, so it needs a decision about where it should point rather than a mechanical fix.

[coveralls]

coverage: 97.976%. remained the same — group_documentation_update into master