Fix pin_backup_t missing the input routing of a pin

[ainyan03]

Fixes #348.

What changes

pin_backup_t::backup() looks for the input signal routed from a pin by reading the input selector (IN_SEL) of the pin's output signal. It compared that selector with the signal number, but IN_SEL holds the GPIO number routed to the signal. So the match only happened when the two numbers were equal by chance, _in_func_num stayed -1, and restore() never put the input routing back.

• Compare with the pin number, as proposed in ESP32: pin_backup_t::backup() が入力選択フィールドをピン番号ではなく信号番号と比較している #348, and only while SIG_IN_SEL is set. With SIG_IN_SEL clear the signal bypasses the GPIO matrix and IN_SEL is a stale value that does not belong to this pin. Matching it would let restore() undo routing that another pin owns.
• Restore order. The input selector write was never reached before, and it sat before the pad registers. It now runs after IO_MUX, GPIO_PIN and the output selector are restored, and before output is re-enabled. This way the input is not reconnected to a pad still configured by the borrower, for example with its input buffer disabled.
• matches_current() now also compares the input selector, so a pin whose input routing was changed while borrowed is reported as changed. Board detection uses it to decide whether a pin needs restoring.

Checked

• Register layout against the ESP-IDF 5.5 headers for ESP32, S2, S3, C2, C3, C5, C6, C61, H2 and P4. IN_SEL holds a GPIO number on all of them. The constant inputs are above the GPIO range. GPIO_SIG0_IN_SEL / GPIO_SIG1_IN_SEL exist on each. On the P4, the GPIO_FUNC1_IN_SEL_CFG_REG based read and the GPIO.func_in_sel_cfg[] write hit the same register.
• Hardware, with the internal I2C bus (M5.In_I2C). The test backs up SDA/SCL, routes their input signals to the constant-high input, calls restore(), and reads IN_SEL again:

Board	Chip	Before	After
Core2 v1.3, Station	ESP32	IN_SEL stays 0x38	back to 21/22
CoreS3	ESP32-S3	IN_SEL stays 0x38	back to 12/11
StopWatch, PaperMono	ESP32-S3	IN_SEL stays 0x38	back to 47/48
Tab5	ESP32-P4	IN_SEL stays 0x3F	back to 31/32

• Board detection result and the resulting display/touch configuration are unchanged before/after on 9 boards: the 6 above plus Basic, ToughC5 and NanoC6.
• Builds with Arduino 2.0.17 / 3.x and ESP-IDF 4.4 to 6.1 across ESP32, S3, C3, C5, C6, C61, H2 and P4.

Not tested

• ESP32-S2, C2 and C3 hardware: register definitions and builds only.