fix: preserve saturated lazy ROC call clocks

[luisleo526]

Outcome

Preserve TradingView's observed saturated lazy-call clock for the two direct ta.roc(close, 3) sites in QuantByBoji without changing ROC globally.

• gives each selected callsite its own copyable clock
• keeps eager close[3] behavior for first call and gap-one/two cases
• uses the previous committed call sample only after a saturated gap of at least three bars
• supports same-bar recomputation, lifecycle reset, and calc-on-order-fill checkpoint/restore
• allocates collision-safe generated names
• selects by AST shape and builtin binding, never slug, generated member number, or comparator polarity

The classifier is deliberately fail-closed. It excludes UDF, request.security, or, ternary, aliased/dynamic source, non-literal/other length, named/dynamic argument, nested control-flow, reassignment, expression-statement, and user-shadowed close shapes.

TradingView evidence

The long sites were pinned by E2/E4/E5. The separately reviewed one-shot S2 short-site oracle is public in pinescript-scrapper PR #29:

• exact 60-row ordinary prefix
• 8 trace trades / 16 payload copies
• reached ROC +0.038753340977, uniquely saturated q1
• eager/q3 absolute errors 1.025336599509 / 1.139529453306
• fail-closed audit PASS

Exact target gate

• engine/TV/strict/price exact: 136/136/136/136
• count absolute delta: 0
• entry/exit/PnL P90: 0 / 0 / 0
• generated C++: e39e7a9fc89c4198e9b2a6b799acd05851d760c001154c4dbfa2071683729bbc
• engine tape: ef65c3fc2d10c1c8754f43a0d2a5912ba4617988b6ca951a10cdfb6d0cd50eee

The rebase onto public codegen 228fe316 preserved stable patch-ID 7de31e6b. Versus the pre-rebase generated Quant output, the only additional base delta is the already-published argument-preserving strategy.close_all("END of DAY", ...); the engine tape remains byte-identical.

Regression gates

• full codegen suite: exit 0, 1,348 collected with 2 existing skips
• focused lazy-ROC compile/runtime and close-all/signature/support suites: PASS
• independent classifier/lifecycle/COOF/name-collision cold review: PASS
• public engine/corpus tree: engine 2dc25ec89711324c8cd028d78680a39ed7c9b389, corpus aeff79e4af1052528010e1b9cded5dd0cc677021
• forced-no-cache scraper sweep: 412/412 accounted (404 ok + 1 no-trades + 6 transpile + 1 existing run error)
• board: 344 Excellent / 50 Strong / 3 Moderate / 5 Weak
• tier movement: QuantByBoji Moderate -> Excellent, 1 UP / 0 DOWN
• final tier report: 03367c5edc5f962311ed691172e75d44e883f604afebd8ac17d0913f265f6845
• explicit pre-patch baseline: dd5708c05880094833dde07769ba6df1e21c0b41b9d7fdb7358c44f02a5bfc9d
• git diff --check: PASS