H.265: reassembled FU NAL units get the FU header as the second NAL header byte #43

Found by the new unit tests (#34).

NalAssembler.assemble() handles the start fragment of an H.265 FU (type 49) like this:

nalBuffer[nalSize++] = (byte)((rx[cp] & 0x81) | ((rx[cp + 2] & 0x7F) << 1));
int donlField = (rx[cp + 1] & 0x80) >> 7;
cp += donlField != 0 ? 3 : 2;

The code writes only the first byte of the 2-byte H.265 NAL header. After cp += 2, copying starts at the FU header, so the FU header becomes the second NAL header byte.

Example: IDR_W_RADL with payload header 62 01 and FU header 93. The result is 00 00 00 01 26 93 ... instead of 00 00 00 01 26 01 .... Decoded, 0x93 means nuh_layer_id = 18 and TemporalId = 2.

Effects:

• H.265 says a single-layer decoder ignores NAL units with nuh_layer_id > 0. Fragmented NAL units, which in practice means every IDR and most P frames, can therefore be dropped by a conforming decoder, leaving a black picture. Decoders that do not check the layer id play the stream.
• The "DONL" check reads a layer-id bit, not a DONL flag. DONL is only present when the SDP has sprop-max-don-diff > 0.
• (fu & 0x7F) << 1 also shifts the E bit into the F bit; it should be & 0x3F.

The basic branch writes 0x01 as the second byte, which is close to correct.

Fix: copy the second byte of the payload header (rx[cp + 1]), mask the type with 0x3F, and skip 3 bytes (payload header + FU header).