ef9345: fix insert and cursor rendering logic

[fabio-d]

Hello, this PR rewrites how the "insert" attribute is processed to solve the issue discussed in jfdelnero/minitel#3 (comment) and briefly mentioned in #13200.

TL;DR: MAME's current handling of the background color is wrong for the TS9347 video chip variant, and the "insert" attribute is handled wrong on EF9345 too. In addition, the hardware cursor should blink twice as fast as flashing text, instead of the same rate.

Longer recap from that discussion:

• EF9345/TS9347 have a per-character attribute called "insert" that, in conjunction with the PAT register, can be used to force colors to black in some conditions.
• The minitel2's demov1 bios rendered correctly in MAME, but ft_bv4 and ft_bv9 display very noticeable glitches in many Videotex pages, such as those in the screenshots here:
  
  instead of
  
• It turned out that MAME currently contains wrong code that always forces the background of all characters rendered by the minitel2 machine to black, regardless of insert, PAT or the character's background color.
  

  mame/src/devices/video/ef9345.cpp

  Lines 417 to 420 in b4d55be

  	if (m_variant == EF9345_MODE::TYPE_TS9347)
  	{
  	c0 = 0;
  	}

• Just removing those lines made ft_bv4 and ft_bv9 render correctly but broke demov1, because MAME's current support for "insert" (on which demov1 relies in order to set black backgrounds) is wrong/incomplete

  mame/src/devices/video/ef9345.cpp

  Lines 433 to 440 in b4d55be

  	if ((m_pat & 0x30) == 0x30)
  	insert = 0; //active area mark
  	if (insert == 0)
  	c1 += 8; //foreground color
  	if ((m_pat & 0x30) == 0x00)
  	insert = 1; //insert mode
  	if (insert == 0)
  	c0 += 8; //background color

  
  

@jfdelnero and I initially made some experiments over the Internet on a real Minitel 2 to better understand how the PAT register works. I've then created a custom "development" board hosting real EF9345/TS9347 chips I bought on eBay. Thanks to this board I've written automatic tests to exhaustively capture the output of all the combinations of PAT and insert attribute values and, in addition, the interactions with the hardware cursor too.

This PR:

• Removes the wrong "background is always black" code block.
• Properly implements the insert attribute under all the possible PAT values (pixel-perfect validated against the output of the real chip).
• Fixes the timing of the hardware cursor which, when positioned on flashing text, shows that it alternates between 4 phases instead of 2:
  • Old MAME rendering (incorrect):
    
  • Correct rendering (with this PR), validated with the real chip:
    
    (test: test_40columns_cursor/pat60mat60a08b00).