[bey0ne]

IR: TIM1 + DMA engine, external IR module and progress bar

1. Play PP4 frames with TIM1 + DMA instead of a busy loop

Problem

tagtinker_ir_transmit() bit-bangs each frame with a DWT->CYCCNT busy loop
inside FURI_CRITICAL_ENTER/EXIT. The comment says the critical section lasts
1-2 ms, but one PP4 symbol takes ~191 us on average (40 us burst + 151 us mean
gap), i.e. ~0.77 ms per byte:

Frame	Interrupts masked
~30-byte data	~23 ms
96 bytes (max)	~74 ms

While masked, USB, BLE, the GUI and the expansion UART stall, and the TX thread
spins at the highest priority between frames.

Change

The stock furi_hal_infrared async TX cannot be reused: it caps the carrier at
INFRARED_MAX_FREQUENCY (1 MHz) and ESLs need ~1.25 MHz. So this PR drives TIM1
directly with the same technique the firmware uses:

• TIM1 counts carrier periods (64 MHz / 51 = 1.255 MHz, unchanged).
• The repetition counter makes TIM1 raise one update event per segment.
• On each update, DMA2 ch1 writes the next output mode to CCMR2
  (PWM2 = carrier on, forced inactive = off) and DMA2 ch2 writes the next
  segment length to RCR.
• The TX thread sleeps on a semaphore until the DMA transfer-complete IRQ.

Files:

• ir/tagtinker_ir_seq.[ch] (new, pure C): builds the DMA tables from a frame.
  No Flipper headers, so it is unit-testable on a PC.
• ir/tagtinker_ir.c: hardware driver. Public API unchanged, no caller changes.
• tests/test_ir_seq.c + make test + a tests CI job (ASan + UBSan):
  replays the tables through a model of TIM1 + DMA and compares the waveform
  with the timing of the old driver (2000 random frames + edge cases).
• README.md / CONTRIBUTING.md: document make test.
• build_and_flash.bat: small Windows helper (ufbt update, build,
  ufbt launch) for contributors without make. Happy to drop it or move it
  to a separate PR if you prefer.

Behaviour notes

• Each segment is rounded to whole carrier periods: every edge is within
  0.4 us of the old timing, every symbol (burst + gap) within 0.8 us. Bursts
  are 40.6 us instead of 40.3 us. PP4 is pulse-position coded, so this is far
  below the ~60 us spacing between symbol values.
• Max frame length is now 96 bytes (TAGTINKER_MAX_FRAME_SIZE) instead of 255.
  All callers already use frames of at most 96 bytes.
• tagtinker_ir_init() refuses to start if furi_hal_infrared_is_busy().
• tagtinker_ir_stop() only signals; the TX thread halts the hardware itself.
• tagtinker_ir_is_busy() was declared but never defined; it is now.
• Uses ~2.3 KB of heap for the DMA tables while the transmit scene is open.

Tested

• make test passes (host, ASan + UBSan); the test fails if two gap constants
  are swapped.
• Cross-compiled for Cortex-M4 with -Wall -Wextra -Werror against the
  firmware headers; all external symbols are exported in api_symbols.csv.
• On hardware: Flipper Zero, firmware mntm 12 (API 88.2), Pricer F14
  (type 1327, 208x112 red):
  • LED test
  • Text
  • Image, page 1
  • Back during a long send stops cleanly
  • USB stays connected during a send (CLI log streamed while sending)
  • Measured frame duration = computed duration within ~1.5 us

2. External IR module on GPIO A7 (#39, #45, #52)

Problem

The built-in IR LED has a short range; several users get partial or no
image updates (#39, #45, #52).

Change

Two settings, like the stock Infrared app:

• IR Output: Internal / GPIO A7 / Auto (detects a module when sending);
• 5V on GPIO: powers modules that take 5V from pin 1.

The driver uses TIM1 CH1N on PA7 instead of CH3N on PB9. Auto uses
furi_hal_infrared_detect_tx_output() and, like the stock app, powers a
detected module. The 5V rail is only switched on if it was off and is
restored afterwards. The transmit screen says "Point the IR module at tag".
Old settings files are still read.

Tested

Flipper Zero, firmware mntm 12, external IR module:
Internal / GPIO A7 / Auto, settings kept after restart, stock Infrared app
fine afterwards, image sent to a Pricer F14 through the module.

3. Progress bar while sending images (#45)

Problem

An image send takes up to ~30 s with only three animated dots (#45).

Change

Image and multi-frame sends draw a bar with a percentage. Each payload
covers a share of the send and each data frame moves the bar. Single
commands and Repeat mode keep the dots. This needs the DMA engine: the old
driver blocked the CPU during frames.

Tested

Flipper Zero, firmware mntm 12: bar from 0 to 100 %, Back during a
send stops cleanly and a new send works.