snapshots: measure SELECT / READ commands and show timing stats

Server measures every classified APDU (A4 select, B0 read binary, B2 read
record) from command to response: _collect_apdu_times() enables the tracer
(reattaching it if pySim nulled it) and /api/select + /api/read return
'apdu_times': [{type, ms}]. Collection is safe: handlers hold _CARD_LOCK.

Snapshots store per-file {select_ms, read_ms}, a ms value per record and
snapshot-level stats {select, read_binary, read_record}: {min, max, avg,
count} plus total_ms (wall time of the scan). The snapshot view gets a
summary under the title (files/records counts, scan time, min/avg/max per
command type) and shows select/read per file and read time per record.
Timings are display-only: checks, snapshots comparison and imports ignore
them (old snapshots simply show 'No timing data').

Tests: Python classifier/collection (tests/test_apdu_timing.py) and
frontend stats/accumulator/format/build-file/summary. SW cache v99 ->
v100; help, README and docs/api.md updated.

frontend/tests/html.test.js

@@ -37,3 +37,7 @@ test('phone simulator has Phone / TR Config pills', () => {
 test('scan name input starts scanning on Enter', () => {
     assert.match(html, /id="profiler-scan-name"[^>]*onkeydown="profilerScanNameKeydown\(event\)"/);
 });
+
+test('snapshot view has a timing summary block', () => {
+    assert.ok(html.includes('id="snapshot-summary"'));
+});