Fix PM1 button and wake IRQ handling and correct the power API docs

[ainyan03]

Problem

A register-level audit of M5PM1_Class against the PM1 datasheet found
several defects:

• getPekPress() reported a double click with the same value as a single
  click, and cleared the whole IRQ Status 3 register on any click — erasing a
  pending WAKEUP flag as a side effect. It also cleared flags that arrived
  between its read and its write.
• clearWakeSource() used a read-modify-write: a wake bit that rose between
  the read and the write was written back as zero and lost. Worse, the read
  path folds an I2C failure into 0, so a single failed read caused the
  write to wipe every wake flag regardless of the mask.
• gpio_function_t exposed wake = 0b10, which the datasheet defines as
  Reserved (the real wake configuration lives in separate registers this
  class does not touch), and setGPIOFunction() accepted any cast value,
  shifting out-of-range encodings into neighboring pins' mux fields.
• getBatteryCharge(bool*) and getBatteryVoltage(uint16_t*) dereferenced
  their output pointers without a null check.
• Several comments were wrong: the DCDC rail is 3.3 V, not 5 V
  (PWR_CFG bit1 = 3.3V_DCDC_EN), and the charge current/voltage/status
  methods documented parameter ranges for registers the PM1 does not have —
  they are permanent stubs.

Fix

• getPekPress() keeps the AXP-shaped contract (0 / 2; the PM1 does not
  report long presses) and consumes the SINGLE and DOUBLE click flags with a
  single selective write-zero-to-clear access that preserves WAKEUP. A double
  click still reports 2 so Power.getKeyState() consumers see a click, and
  the new wasPekDoubleClicked() reports the distinction without extra I2C
  traffic. If the clear write fails the event stays pending and 0 is
  returned instead of consuming it.
• clearWakeSource() performs one selective write with no read, masked to
  the documented bits [6:0]. The docs now state the write-zero-to-clear
  assumption explicitly, since the datasheet does not specify the polarity.
• gpio_function_t::wake is removed and setGPIOFunction() rejects
  anything but gpio, irq, and special.
• Null guards added to the two output-pointer getters, matching the
  neighboring APIs.
• Doc corrections for the 3.3 V DCDC rail and the charge-control stubs; the
  dead commented-out implementations referencing nonexistent registers are
  removed.

Breaking change

gpio_function_t::wake no longer exists; code naming it fails to compile.
It never worked — it wrote a reserved encoding.

Post-review amendment

A second local adversarial review round (two independent passes over the
final diff) converged on one real defect: the double-click latch was
set-only, so a stale double could be attributed to a much later single
click. The latch is now reassigned on every successfully reported click.
The wasPekDoubleClicked() doc also states the single-task polling
expectation, and the charge getter docs no longer show value ranges for
registers that do not exist.