Add warmup=False option to Strategy.I() to exclude an indicator from forced backtest warm-up (#1339)

[Harsh4r0ra]

Problem

Backtest.run() currently determines the first bar on which Strategy.next()
is called by taking the max over all declared indicators of the index
of their first non-NaN value (_indicator_warmup_nbars() in _util.py).
In other words, one slow-to-warm-up indicator delays the whole backtest,
even if that indicator isn't actually needed for trading decisions at that
point.

This becomes a real problem when mixing timeframes with
backtesting.lib.resample_apply(). As reported in #1339, a strategy running
on 30-minute bars that also computes a 100-period SMA on the resampled
daily timeframe ends up with ~100 days worth of NaNs at the base
resolution — i.e. thousands of 30-min bars — before the backtest places a
single order, even though the strategy's actual entry/exit logic only
depends on fast, lower-timeframe indicators. Users currently have no way to
tell backtesting.py "this indicator is secondary/informational, don't let
its warm-up gate the whole run."

Fix

Added a warmup: bool = True keyword argument to Strategy.I():

def init(self):
    self.sma_fast = self.I(SMA, self.data.Close, 5)               # gates warm-up (default)
    self.sma_slow_1d = resample_apply('1D', SMA, self.data.Close, 100,
                                       warmup=False)                # excluded from warm-up

Summary

• Fixes backtesting library skips bars where any indicator has NA values (or don't place orders) #1339 — currently the backtest is forced to skip bars until every
  declared indicator has non-NaN values, which is painful for indicators
  computed on a higher/resampled time frame via resample_apply (their
  warm-up in base-timeframe bars can be huge) when those indicators aren't
  actually required from the very first bar.
• Adds a warmup: bool = True parameter to Strategy.I(). Indicators
  declared with warmup=False are excluded from the automatic warm-up-bar
  calculation in _indicator_warmup_nbars, so the backtest can start as soon
  as its required indicators are ready.
• The option flows through automatically to resample_apply(..., warmup=False)
  since it forwards unrecognized kwargs to Strategy.I().

Test plan

• Added TestStrategy.test_indicator_warmup_false_excluded_from_start_bar
• Full suite passes (pytest backtesting/test/_test.py, 81 passed)
• flake8 clean