Fix STC dividing by a near-zero denominator, producing -Inf/huge out-of-range swings

[cinar]

Closes #425.

Summary

• trend.Stc.ComputeWithContext divided (MACD-%K)/(%D-%K) for its final line, per the doc comment's STC = 100 * (MACD - %K) / (%D - %K). %D-%K is frequently near zero (both are stochastic-smoothed values of the same MACD series and track each other closely), so real-world input yields -Inf and swings like 1435/-710 for an indicator documented to oscillate 0-100.
• Replaced with the standard Schaff Trend Cycle algorithm: apply the existing Stochastic (rolling-min/max normalization + SMA) twice -- once to MACD to get %K1/%D1, then again to %D1 to get %K2/%D2 -- with STC = %D2. Both stages stay bounded in [0, 100] by construction.
• %K1/%K2 are discarded (STC only needs the doubly-smoothed %D), but each comes from the same internal duplicate fan-out as %D1/%D2 inside Stochastic.ComputeWithContext, so each is drained in its own goroutine (helper.DrainWithContext) -- an unread duplicate branch otherwise blocks the shared producer and stalls %D1/%D2 too (this reintroduces the class of deadlock Fix Stc.ComputeWithContext output channel never closing #421 fixed if skipped).
• IdlePeriod() updated to Apo.IdlePeriod() + 2*Stochastic.IdlePeriod() to match the now-doubled warm-up.
• TestStcFull's fixture (testdata/stochastic.csv, 19 rows, shared with two other tests) is too short for the new warm-up, so it now generates its own longer synthetic series instead.

Note on remaining NaN

STC can still legitimately output NaN when a stochastic stage's rolling window is exactly flat (a genuine 0/0). Separately, I found that trend.MovingSum (which the Sma inside Stochastic uses) doesn't recover from a single NaN even after it leaves the window -- filed and fixed separately as #427 / PR (fix/movingsum-nan-poisoning), since it's a shared primitive bug, not specific to STC. With that fix applied too, STC on a 251-day BRK-B series drops from 157/180 NaN outputs down to 36/180, with the rest being either the initial warm-up or a genuinely flat window at the very tail (no future data to recover into). Without it, STC's math is still correct here, just noisier in practice on real data until #427 lands.

Test plan

• go build ./...
• go test -race ./... (full repo, no regressions)
• go vet, gosec, staticcheck, revive clean on trend/
• Manual check against asset/testdata/repository/brk-b.csv: output now bounded in [0, 100] (previously -Inf, 1435, -710, etc.)
• New TestStcFull (longer synthetic fixture) and existing TestStcDeadlock/TestStcSlowStochastic all pass

🤖 Generated with Claude Code

[cinar]

Depends on #430 (MovingSum NaN-poisoning fix) for STC to be fully usable on real data -- STC's math is correct here regardless, but a single flat MACD window anywhere in the series will otherwise cause an extended NaN run downstream until #430 also lands. See #430's description for the before/after numbers.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.29%. Comparing base (6653727) to head (9022386).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #428      +/-   ##
==========================================
- Coverage   92.32%   92.29%   -0.03%     
==========================================
  Files         229      229              
  Lines        7280     7259      -21     
==========================================
- Hits         6721     6700      -21     
  Misses        472      472              
  Partials       87       87              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.