Add 32-bit float register decoding (--float)

[19bk]

Many industrial sensors store values as 32-bit IEEE 754 floats across two consecutive registers. Add a `--float` flag that reads register pairs and decodes them:

```bash
modbus read 192.168.1.10 40001 -c 4 --float

Reads 4 registers, decodes as 2 float values

```

Should support both big-endian and little-endian byte/word order (`--byte-order BE` / `--word-order LE`), since different manufacturers use different conventions.

Python struct module handles the actual decoding. The tricky part is the byte order options.

[passionworkeer]

For --float, I would keep decoding as a pure transformation layer after the raw register read so transport/read logic stays unchanged. A simple pattern is: group registers in pairs, convert each 16-bit register to two bytes, then apply byte-order and word-order separately before struct.unpack. That makes it easy to support common layouts like BE/BE, BE/LE, LE/BE, LE/LE and to add table-driven tests with known register pairs -> expected float values. Also worth failing fast when the register count is odd.

[zakiscoding]

I'd like to work on this! I'll add a \ flag with \ and \ support for 32-bit IEEE 754 float decoding and submit a PR shortly.

[19bk]

Really like this framing @passionworkeer — keeping float decoding as a pure transformation layer after the raw read is clean. The byte-order/word-order matrix (BE/BE, BE/LE, LE/BE, LE/LE) is a must for real-world Modbus devices since every manufacturer seems to pick a different combo. And yeah, failing fast on odd register counts will save people a confusing debug session.

@zakiscoding PR #6 is up for this — looking forward to reviewing it. Let's make sure we cover those byte-order combos.