power: supply: bq25890: Fix race causing oops at boot

Before this commit the driver was registering its interrupt handler before
it registered the power_supply, causing bq->charger to potentially be NULL
when the interrupt handler runs, triggering a NULL pointer exception in
the interrupt handler:

[   21.213531] BUG: kernel NULL pointer dereference, address: 0000000000000680
...
[   21.213573] Hardware name: Xiaomi Inc Mipad2/Mipad, BIOS MIPad-P4.X64.0043.R03.1603071414 03/07/2016
[   21.213576] RIP: 0010:__lock_acquire+0x5c5/0x1de0
...
[   21.213629] Call Trace:
[   21.213636]  ? disable_irq_nosync+0x10/0x10
[   21.213644]  ? __mutex_unlock_slowpath+0x35/0x260
[   21.213655]  lock_acquire+0xb5/0x2b0
[   21.213661]  ? power_supply_changed+0x23/0x90
[   21.213670]  ? disable_irq_nosync+0x10/0x10
[   21.213676]  _raw_spin_lock_irqsave+0x48/0x60
[   21.213682]  ? power_supply_changed+0x23/0x90
[   21.213687]  power_supply_changed+0x23/0x90
[   21.213697]  __bq25890_handle_irq+0x5e/0xe0 [bq25890_charger]
[   21.213709]  bq25890_irq_handler_thread+0x26/0x40 [bq25890_charger]
[   21.213718]  irq_thread_fn+0x20/0x60
...

Fix this by moving the power_supply_register() call to above the
request_threaded_irq() call.

Note this fix includes making the following 2 (necessary) changes:

1. Switch to the devm version of power_supply_register() to avoid the
need to make the error-handling in probe() more complicated.

2. Rename the "irq_fail" label to "err_unregister_usb_notifier" since
the old name no longer makes sense after this fix.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>